J Forensic Sci, November 2013, Vol. 58, No.

6
doi: 10.1111/1556-4029.12203
Available online at: onlinelibrary.wiley.com

PAPER
GENERAL

Matthew B. Thompson,1,2 B.Inf.Tech., B.Sc. (Hons); Jason M. Tangen,1 Ph.D.; and

Duncan J. McCarthy,3 B.App.Sc.

Expertise in Fingerprint Identification*

ABSTRACT: Although fingerprint experts have presented evidence in criminal courts for more than a century, there have been few scientific
investigations of the human capacity to discriminate these patterns. A recent latent print matching experiment shows that qualified, court-practicing fingerprint experts are exceedingly accurate (and more conservative) compared with novices, but they do make errors. Here, a rationale
for the design of this experiment is provided. We argue that fidelity, generalizability, and control must be balanced to answer important
research questions; that the proficiency and competence of fingerprint examiners are best determined when experiments include highly similar
print pairs, in a signal detection paradigm, where the ground truth is known; and that inferring from this experiment the statement The error
rate of fingerprint identification is 0.68% would be unjustified. In closing, the ramifications of these findings for the future psychological study
of forensic expertise and the implications for expert testimony and public policy are considered.

KEYWORDS: forensic science, fingerprints, decision making, expertise, testimony, judgment, law, policy

Maintaining high standards of forensic evidence is vital for an

effective justice system and for ensuring that innocent people are
not wrongfully accused. Although fingerprint experts have presented evidence in criminal courts for more than a century, there
have been few scientific investigations of the human capacity to
discriminate these patterns and impressions. Contrary to popular
belief (and television shows like CSI), computers are not relied
upon to match crime scene fingerprints. Instead, human fingerprint experts decide whether a print belongs to a suspect or not.
These experts make thousands of fingerprint identifications, per
day, to be used as evidence in courts of law. Until recently, it
was unclear what role expertise plays or whether expertise is
even necessary to conduct accurate fingerprint comparisons.
In 2011, we (Tangen, Thompson, and McCarthy) published
the results of an experiment testing the accuracy and claimed
expertise of fingerprint examiners (1). These results showed that
qualified, court-practicing fingerprint experts are exceedingly
accurate (and more conservative) compared with novices, but
they do make errors. Here, the current state of fingerprint testimony, measures of accuracy, and the research culture in forensic
science are discussed. A rationale for the Identifying Fingerprint Expertise (1) experimental design is provided, and the
steps taken to balance fidelity, generalizability, and control;
ensure validity and ground truth; create a signal detection
1
School of Psychology, The University of Queensland, St Lucia, QLD
4072, Australia.
2
Queensland Research Laboratory, National Information and Communications Technology Australia, St Lucia, QLD 4072, Australia.
3
Forensic Services Branch, Queensland Police Service, 200 Roma St, Brisbane, QLD 4000, Australia.
*Supported by a Fulbright Scholarship to Thompson, and an Australian
Research Council Linkage Grant to Tangen and McCarthy (see The Forensic
Reasoning Project at ForensicReasoning.com).
Received 2 Dec. 2011; and in revised form 31 July 2012; accepted 11
Aug. 2012.

2013 American Academy of Forensic Sciences

framework with highly similar prints; establish expertise with a

novice control group; and establish meaningful error rates are
described. Given the brevity of the original research article, this
rationale will provide context for interpreting the results for the
benefit of researchers, forensic examiners, forensic managers,
lawyers, and judges. In closing, the ramifications of the findings
for the future study of forensic expertise and the implications for
expert testimony and public policy are considered.
Fingerprint Expert Testimony
In 2009, the National Academy of Sciences (NAS) delivered a
landmark report highlighting the absence of solid scientific methods and practices in the forensic science domain (2). Harry T.
Edwards (a senior U.S. judge and co-chair of the NAS Committee) noted that forensic science disciplines, including fingerprint
comparison, are typically not grounded in scientific methodology, and forensic experts do not follow scientifically rigorous
procedures for interpretation that ensure that the forensic evidence offered in court is valid and reliable [(3); see also (47)].
The NAS report (2) highlighted the absence of experiments on
human expertise in forensic pattern matching: The simple reality is that the interpretation of forensic evidence is not always
based on scientific studies to determine its validity. This is a
serious problem (p. 8). They recommended that the U.S. Congress fund basic research to help the forensic community
strengthen their field, develop valid and reliable measures of performance, and establish evidence-based standards for analyzing
and reporting testimony.
Courts rely heavily on forensic evidence to convict the guilty
and protect the innocent. The presentation of flawed forensic
evidence has obvious implications for individual cases, but it
also calls into question the integrity of the entire criminal justice
systeminnocent people may be wrongfully convicted and people may lose trust in the justice system (3). It is important, there1519

1520

JOURNAL OF FORENSIC SCIENCES

fore, that the claims made by fingerprint examiners testifying in

court are accurate, substantiated, and reasonable. Fingerprint
examiners have claimed that fingerprint identification is infallible
(8) and that there is a zero error rate for fingerprint comparisons
(3,9). Several commentators, however, have suggested that the
claims of individualization and a zero error rate are not supported by evidence and, moreover, are scientifically implausible
(e.g., [2,10]). Past President of the International Association for
Identification suggested that members not assert 100% infallibility (zero error rate) of fingerprint comparisons (11) and the Scientific Working Group on Friction Ridge Analysis, Study and
Technology (SWGFAST) has drafted a standard for defining,
calculating, and reporting error rates (12).
These issues are made more complicated by the nature of the
legal system. The adversarial approach to the submission of evidence in court is not well suited to establishing scientific truth.
According to Edwards (3), judges and lawyers generally lack the
expertise necessary to evaluate forensic evidence scientifically;
defense attorneys often lack the resources to challenge the evidence; judges make admissibility decisions without the benefit
of judicial colleagues and time to research; and cases are seldom
appealed on the basis of disputed forensic evidence. It may be
unwise, therefore, to rely on the judicial system to address the
challenges facing fingerprint expert testimony.
Proficiency Tests and Accuracy
Fingerprint proficiency tests are availablesuch as those provided by Collaborative Testing Services, Inc. (CTS)where the
goal is to measure the accuracy of participating laboratories as a
unit. The results are sometimes reported as a function of the
kind of source print, sometimes as a function of the kind of correct response, sometimes as a function of the kind of error, and
sometimes as the proportion of examiners/laboratories producing
various responses. Proficiency tests of fingerprint examiners and
previous studies of examiners performance have been heavily
criticized for (among other things) failing to include large, counterbalanced samples of targets and distractors for which the
ground truth is known (see [1315]). A weakness of proficiency
tests and previous experiments is that a particular crime scene
(or latent) print either forms part of a match comparison or
part of a distractor comparison for every individual who takes
the testa particular latent never serves as part of a target trial
for one examiner/laboratory and a distractor trial for another
examiner/laboratory. The result is that even a single highly distinctive latent on a distractor trial can artificially improve discrimination by reducing false positives. Or a single highly
distinctive latent in a match trial can artificially improve discrimination by increasing hits (15).
There is nothing inherently wrong with the proficiency tests,
like those provided by CTS, if the goal is to measure examiners
performance on exactly the same set of items. It may be possible
to narrow in on particular features that cause difficulty (e.g., a
peculiar pattern type) or what prints a particular department has
trouble with. Indeed, if CTS made their materials and all their
results widely available, they may provide a useful tool to measure performance on specific items and for assessing reliability.
But, the tests are insufficient for measuring accuracy. To make
general claimsbeyond those of specific prints at a specific
level (e.g., accuracy with whorl patterns)different (and randomized) sets of prints for each examiner are needed. Otherwise,
information in the specific prints used for the test will influence
performance, making it difficult to generalize the results (15).

Proficiency tests have not adequately addressed the general issue

of expert matching accuracy and are not designed to disentangle
the factors that affect matching accuracy.
There is, however, a growing body of research on fingerprint
matching. Researchers have investigated the effect of contextual
bias on fingerprint examiners (e.g., [1621]); some of the special
abilities and vulnerabilities of fingerprint examiners (e.g.,
[2225]); the effect of technology (e.g., [26,27]); statistical
models of fingerprint identification (e.g., [2831]); and the
accuracy of fingerprint examiners decisions (e.g., (3236) but
see [15,37,38]). But, despite its 100 year history, there have still
been few peer-reviewed studies directly examining the extent to
which experts can correctly match fingerprints to one another,
how competent and proficient fingerprint experts are, how and
on what basis examiners make their decisions, or the factors that
affect matching accuracy and what is the effect of expertise. In
this study, we focus our efforts on the claimed matching
expertise of fingerprint examiners.
The Research Culture
There is little doubt, among critics and proponents alike, that
fingerprint identification is a valuable tool for law enforcement.
Fingerprint identification errors are unlikely to be made because
of malicious actionsfingerprint experts do their best to provide
accurate fingerprint evidence to the courts and uphold civil liberties. Indeed, in the wake of the Mayfield case of false identification, the FBI stated their intention to make certain they are
employing the most effective means to ensure the integrity of
their expert fingerprint examinations (39). Butunlike other
areas of expertise where decisions are safety-critical, such as aviation and medicinethere is currently no culture of research in
fingerprint identification (40). Steady advances in the fingerprint
development process have been made, but the critical human
decision-making element has been neglected. Examiners are
eager to demonstrate their abilities and advance their field, but
rarely receive the support and resources to do so. The Director
of the FBIs Investigation Lab described the gap between basic
research and its application in solving crimes as the valley of
death because nobody wants to pay for it, nobody really wants
to do it (41).
It appears that fingerprint examiners are expected to
strengthen the scientific basis of their field, while they relentlessly make identifications, search databases, and testify in court.
Examiners, however, do not have the time, infrastructure, training, expertise, or research culture necessary to mount studies of
human performance to ensure their field meets scientific and
legal standards of evidence. By analogy, it would be like expecting the local doctor to find a cure for cancer. Clearly, examiners
are not well positioned to address the challenges leveled at their
field alone and, traditionally, there has not been a good working
relationship between examiners and researchers. Research on
expertise and complex systems is the domain of Cognitive Science and of Human Factors. These researchers have the reward
structures already in place for conducting and publishing highquality research and are well positioned to work with examiners
to strengthen the field.
Much of the existing research on the cognitive factors
involved in fingerprint judgment has investigated the influence
of contextual information on examiners performance. Dror et al.
(17) used a highly publicized case of exposed fingerprint error
to determine whether biasing information could lead an examiner
to change their prior judgment. They covertly evaluated five

THOMPSON ET AL.

examiners, with an average of 17 years of experience, who consented to being tested at an unknown time over 12 months. The
five examiners were each given a print to identify by a colleague, who advised them that the fingerprints were from a
famous case of misidentification by the FBI for the 2004 Madrid
train bombings. One examiner reported that the prints matched,
three reported that the prints did not match, and one reported
inconclusive. Unbeknownst to the examiners, however, the prints
that they were asked to identify were taken from their own previous case history where they had previously declared them a
match. With four of the five examiners subsequently changing
their previous judgment of the prints as matching, it seems that
top-down, contextual influences can affect expert judgments (see
also [16,19]). Dror and Rosenthal (20) also conducted a metaanalysis to determine the degree to which examiners would
make the same or conflicting decisions if extraneous information
about the case was added. Although good data were sparse, the
authors concluded that examiners are susceptible to bias.
It is clear that fingerprint experts have special abilities, but
their decisions can be influenced by extraneous contextual information (22,42), and researchers have suggested ways contextual
bias can be mitigated (43). Even with this contribution, relatively
little research on human fingerprint identification has been conducted by academics and professionals alike. The U.S. NAS (2)
and others have called for the development of a research culture
within forensic science. Mnookin et al. (40) argue that there is a
legitimate role for experience-based claims of knowledge, but
also that pattern identification disciplines must develop a scientific foundation, through research, that is grounded in the values
of empiricism and skepticism. The experiment described below
is a step toward addressing the call from the NAS for the urgent
development of objective measures of accuracy and expertise in
fingerprint identification.
The Identifying Fingerprint Expertise Experiment
The Identifying Fingerprint Expertise experiment (1) was
designed to find out whether fingerprint experts were any more
accurate at matching prints than the average person and to get
an idea of how often they make errors of the sort that could
lead to a failure to identify a criminal compared with how often
they make errors of the sort that could lead to inaccurate evidence being presented in court. Thirty-seven qualified fingerprint experts and 37 undergraduate students were given pairs of
fingerprints to examine and decide whether a simulated crime
scene print matched a potential suspect or not. Some of the
print pairs matched, while others were highly similar but did
not match.
Thirty-six simulated latent crime scene prints were paired with
fully rolled exemplar prints. Across participants, each simulated
print was paired with a fully rolled print from the same individual (match), with a nonmatching but similar exemplar (similar
distractor), and with a random nonmatching exemplar (nonsimilar distractor). The simulated prints and their matches were from
our Forensic Informatics Biometric Repository, so, unlike genuine crime scene prints, they had a known true origin (13,44). All
fingerprints were authentic (i.e., not simulated), but the latents
were simulated in the sense of representing those found at
crime scenes during casework. (For human matching performance with genuine crime scene prints, see [45].) Similar
distractors were obtained by searching the Australian National
Automated Fingerprint Identification System (NAFIS). For each
simulated print, the most highly ranked nonmatching exemplar

FINGERPRINT EXPERTISE

1521

from the search was used if it was available in the Queensland

Police 10-print hard-copy archives, which contains approximately one million 10-print cards (10 million individual prints)
from approximately 300,000400,000 people (one person may
have more than one 10-print card on record).
The results were striking. Of the prints that actually matched,
the experts correctly declared 92.12% of them as matching (hits).
Of the prints that did not actually match, the experts incorrectly
declared 0.68% of them as matching (false alarms)impressive
expert performance, considering the corresponding false alarm
rate for novices was 55.18%. We concluded that qualified courtpracticing fingerprint experts are exceedingly accurate compared
with novices, but are not infallible. Our experts showed a conservative response bias, tending to err on the side of caution by
making more errors of the sort that could allow a guilty person
to escape detection than errors of the sort that could falsely incriminate an innocent person. Even so, they made the kind of error
that could result in incorrect evidence being presented to the
court in a criminal trial.
Balancing Fidelity, Generalizability, and Control
Readers will react differently to the methodology employed in
the Identifying Fingerprint Expertise experiment. The intuition
that the best experiment should resemble real-life as much as
possible is understandable. However, this intuition is incorrect
(see [46] for a defense of external invalidity), as has been demonstrated in other complex, safety-critical domains, such as aviation and medicine. When designing studies of human
performance, the challenge is to find the appropriate balance
between fidelity, generalizability, and control, to produce data
that are best suited to answering the research question (47,48).
Fidelity is the similarity of an experimental task to its reference domain (i.e., fingerprint identification; [49,50]). How well
does the task represent the particular work domain? Is the expertise of the participants high or low? Are the experimental situations full-featured or simplified? Are the available tools
restricted or complete?
Generalizability is the theoretical depth or breadth of applicability of the results to situations beyond those examined in the
study. Can the results and conclusions of the experiment be
extended to situations that are different from the experiment, and
is this the goal?
Control is the latitude available to the experimenters to isolate
and manipulate variables. Is control high or low, and will the
data collected be sensitive enough to detect differences between
the experimental manipulations?
The perfect, but unattainable, experiment, will have high fidelity, high control, and high generalizability. But, these variables
must be balanced to answer the research question appropriately
(51,52). To understand accuracy rates in fingerprint identification, it is tempting to think that the best option is to insert test
printsunbeknownst to the examinersinto their regular workflow and measure the number of errors that come out the other
end. This arrangement of high fidelity comes at the cost of
reduced generalizability (we cannot apply the results from one
experiment in a particular laboratory to fingerprint identification
more broadly) and reduced control (when errors occur, we have
no way of knowing how they arose and, therefore, what we
might do to prevent them).
Our goal in measuring fingerprint expertise was to compare
expert and novice performance at matching fingerprints and to
get an idea of how often they fail to declare matching prints as

1522

JOURNAL OF FORENSIC SCIENCES

such (misses) in a matching task compared with how often they

declare that prints match when they actually do not (false
alarms). In designing the experiment, fidelity, generalizability,
and control were balanced to answer these questions. The goal
was not to generalize the results from these laboratory-based
experiments to the real world (46). The fingerprint examiners
who participated in this experiment did not have their usual tools
available to allow them to zoom, rotate, or apply filters to the
images; the latent prints that were used were collected as part of
the Forensic Informatics Biometric Repository; and examiners
conducted the experiment during their break on laptops that were
provided in bureau conference rooms. This situation is not
purposefullyanalogous to casework. The majority of expert
participants, nonetheless, reported that the task represented their
day-to-day work.
Unlike opinion polls and surveys, scientific experiments are
not about seeing how well a sample approximates the general
population from which it was selected. Laboratory-based experiments are intentionally artificial, because they allow for the control of all factors that are not of interest (e.g., the benefit of
software tools, the role of verification, the type of crime, inconclusive responses, lifting agents, etc.) and for the systematic
manipulation of only the factors of interest (e.g., the difference
between novices and qualified experts, comparing performance
on match trials and nonmatch trials, ensuring the ground truth of
the prints, using highly similar distractors from a national database search, etc.). Generalizability in this context refers to the
extent to which the difference between expert and novice performance is real, not the extent to which the laboratory setting
resembles the everyday operations of a fingerprint bureau.
The unit of analysis in this particular experiment is the comparison between matching and nonmatching prints or between
experts and novices, not their absolute performance. So, even
though a false alarm rate for experts of 0.68% is impressive in
its own right, we cannot determine from this experiment whether
this rate reflects the false alarm rate of the field more generally.
We can conclude, however, that a false alarm rate of 55.18% for
novices pales in comparison with experts under the same conditions. These results provide sufficient evidence for examiners to
legitimately claim specialized knowledge, which may satisfy
legal admissibility criteria. These results do not allow one to
conclude that 0.68% is the misidentification rate for the field. A
full scale, end-to-end black box experiment would allow us to
pinpoint the precise rate of accuracy in a particular manifestation
of practice, but it is inappropriate and inefficient to conduct a
black box experiment to make simple claims like, Are experts
better than novices? and Do experts make errors? Much more
will be said about black box experiments in the Determining
Error Rates section below. Put simply, the design of an experiment needs to be targeted specifically at the question that one
sets out to address.
Validity and Ground Truth
Validity is a cornerstone of the scientific method. It is a measure of whether a method, instrument, questionnaire, construct,
etc., measures what it is supposed to measure. Validity can be
demonstrated by comparing the outcomes of a method with the
ground truth. So, to assess validity in fingerprint identification
decision making, the conclusion of the identification process
(i.e., match or nonmatch) can be compared with that which is
known (i.e., the ground truth). For example, if the ground truth
of a pair of prints is that they were left by two different individ-

uals, but the examiner incorrectly declares that the prints match,
then the examiner has made a false alarm type of error; if the
ground truth of a pair of prints is that they were left by the same
finger from the same individual, but the examiner concludes that
the prints do not match, then the examiner has made a miss
type of error. The same goes for the two ways the examiner can
reach the correct conclusion: that the prints actually match and
the examiner correctly declares them as such (a hit) or that the
prints do not actually match and the examiner correctly declares
them a nonmatch (a correct rejection).
Most tests of proficiency and studies of accuracy (with the
exception of [32]) used print pairs from casework where the
ground truth was uncertain (see [1315,53]). In assessing validitythe degree to which a test measures what it is supposed to
measurein fingerprint identification decision making, experiments ought to use print pairs for which the ground truth is
known.
To make use of ground truth prints in our expertise study, fingerprint pairs were sourced from the Forensic Informatics Biometric Repositoryan open biometric repository that we created
to increase the availability of high-quality forensic materials
where the ground truth is known. Details on the Forensic Informatics Biometric Repository are available at www.FIB-R.com.
Forensic Informatics Biometric Repository contains a range of
crime related materials: fingerprints, palm prints, shoe prints,
face photographs, handwriting samples, voice samples, and iris
photographs. The materials are collected from participants using
a standardized methodology and vary systematically in quality.
The repository contains multiple types of materials converging
on a single source, and the ground truth of the source is built
into the system. Materials are also collected from participants
over two sessions to approximate the natural variation that is
commonly found in forensic evidence (e.g., changes in facial
hair, clothes, and shoe decay). Participants are first-year undergraduates who participate in 1 h of data collection for course
credit and who provide informed consent.
The fingerprint materials contained in Forensic Informatics
Biometric Repository are 10-prints, palm prints, and latent prints.
Ink is used to capture each fingerprint onto standard 10-print
cards, rolled fully from nail-edge to nail-edge, as well as slap
impressions (pressing, not rolling, the fingers on the card) and
fully rolled palms. Latent prints are taken from common crime
scene surfaces (determined in consultation with fingerprint examiners) including: gloss-painted timber, smooth metal, glass, and
smooth plastic. Participants are instructed to interact with the
surfaces by pushing on the gloss-painted timber to open the
door or safely grabbing the knife by the blade. By interacting
with objects in this way, the aim is to approximate the variation
in materials that are commonly found at actual crime scenes.
In our experiment, the latent prints used were mated with their
matching fully rolled exemplar so that the ground truth of match
trials was known. The use of ground truth print pairs means that
we can compare participants responses to reality.
Signal Detection
A signal detection framework was used to measure the matching performance of fingerprint examiners (see also [15,54]). Signal
detection is a method of quantitating a persons ability to distinguish signal from noise. In fingerprint identification, for example, signal refers to print pairs that truly match and noise refers
to print pairs that do not truly match. Signal detection was initially applied to radar operators who were trying to discriminate

THOMPSON ET AL.

friendly and enemy aircraft and has since been used to measure
all areas of human performance. Several factors may affect a
persons ability to distinguish signal from noise, such as experience, expectations, context, physiological, and psychological
states. To conduct a signal detection analysis of novice and
expert fingerprint matching performance, the two ways of being
right and the two ways of being wrong were separated; performance on matching and nonmatching prints was compared; and
accuracy and response bias were separated.
Separate the Two Ways of Being Right and the Two Ways of
Being Wrong
When an examiner compares two fingerprints, there are two
ways for her to be right and two ways to be wrong, as shown in
Fig. 1. To get a comparison right, she can correctly say the
prints match when they actually do (a hit) or she can correctly
say they do not match when they do not (a correct rejection).
These decisions could result in providing correct evidence to the
court or help eliminate potential suspects. To get a comparison
wrong, she can incorrectly say that the prints match when they
do not (a false alarm), or she can incorrectly say that the prints
do not match when in fact they do (a miss). These decisions
could lead to providing incorrect evidence to the court or a failure to identify a criminal.
Compare Performance on Matching and Nonmatching Prints
To properly measure performance, examiners must compare
both matching and nonmatching prints. As discussed in the section on Proficiency Tests and Accuracy above, most previous
studies have included no or few distractors, making it impossible
to measure the two ways of being right and the two ways of
being wrong, leading to artificially inflated accuracy rates.
In this experiment, to avoid the problem of distractors, each
latent print in the set formed part of a match, similar distractor,
and nonsimilar distractor trial. (The reasoning for providing similar distractors is described in the Similarity section below.) This
way, match trials can be directly compared with the same number of nonmatch trials in the other two conditions. For each participant, each latent print was randomly allocated to one of the
three trial types, with the constraint that there were 12 prints in
each condition. This way, each latent print has an equal chance
to act in either a match, similar distractor or nonsimilar distractor
trial, and so eliminating the possibility that a particularly easy/
Fingerprint
Status
Match

Non-Match

Match

Hit

False
Alarm

No
Match

Miss

Correct
Rejection

Examiner
Says

FIG. 1A 2 9 2 contingency table depicting the four possible outcomes

of a forced choice fingerprint matching task where two prints match or not
and an examiner declares them as a match or no match.

FINGERPRINT EXPERTISE

1523

difficult/distinctive/high-quality latent print could artificially

influence examiners performance.
Separate Accuracy and Response Bias
There are two distinct measures of performance in a fingerprint
comparison task. The obvious one is accuracyan examiners
ability to discriminate matches from nonmatches. The less obvious measure is response biasthe decision rule employed by an
examiner when they are uncertain about a comparison. That is,
their tendency to say match or no match regardless of
whether the prints match or not. If an examiner is unsure whether
two prints match, and declares that they do, then they have made
a liberal decision. If an examiner is unsure whether two prints
match, and declares that they do not, then they have made a
conservative decision. Averaged across several comparisons,
the criteria used to make these decisions add up to so-called liberal and conservative response biases. Put simply, a response bias
is a measure of a persons willingness to say, yes: if they say
yes a lot, then they have a liberal response bias; if they say
no a lot, then they have a conservative response bias.
Two examiners can be equally accurate in their ability to discriminate or see matching prints, butif they have a different
response biasthey may come to opposite conclusions. It follows that there is no universal best response bias. There is an
optimal criterion that minimizes false alarms and misses, but the
appropriate decision criterion will depend on the costs and benefits of committing both types of error and both types of success.
Only when the number of response alternatives is limited can
an examiners response bias be separated from their ability to
discriminate prints.
Forcing a choice is a widely used paradigm for measuring
human performance. In our experiment, participants were asked
to judge whether print pairs matched, using a confidence rating
scale ranging from 1 (sure different) to 12 (sure same)
anchored at the center (i.e., 6.5). The response scale forced a
no match or match decision because ratings of 1 through 6
indicated no match, whereas ratings of 7 through 12 indicated a
match. (Note that these ratings were described incorrectly in the
Procedure section of our original paper.) That is, subjects were
required to move the scrollbar either left (to six or less, different) or right (to seven or more, same); they could not make a
rating of 6.5. This 12-point confidence scale was not designed to
reflect the decisions made, and terms used, by examiners during
casework. Judgments that the information was inconclusive,
which are often made in practice, were not permitted in this
match/no match forced-choice design, making it possible to distinguish between accuracy and response bias (55). Interestingly,
experts responded much more toward the extreme ends of the
scale compared with novices: 92% of expert responses were
either a 1 or a 12 compared with 32% for novices.
Aside from the capacity of the forced-choice procedure to differentiate the roles of accuracy and response bias, there are difficulties with measuring inconclusive judgments. There is no
ground truth for sufficiency, that is, there is no way of knowing
whether a print contains sufficient information for a human to
discriminate it. The best that can be done is to ask several experts
about the sufficiency of the information in several prints to see
whether they agree with each other and themselves on repeated
examinations (i.e., between and within participant reliability).
What one means by insufficient is also tricky. Does it mean,
There is not sufficient information in the latent print to make an
identification, or does it mean, I am unwilling to make a

1524

JOURNAL OF FORENSIC SCIENCES

judgment (match/identification, no match/exclusion) either way.

In fact, if a sufficient amount of information or signal was present
(whatever that means), and an examiner declared it inconclusive,
then this ought to be regarded as a miss type of error.
Sufficiency of information in this experiment was partially
controlled by only using prints that an expert declared as having
sufficient information to make an identification. Participants
uncertainty in their judgments was also controlled using a
12-point confidence scale where a rating of 6, for example,
would be counted as a nonmatch decision.

the proprietary nature of NAFIS, the information it uses in its

search algorithms is unknown; but, it almost certainly relies on the
minutiae, features, direction, relative and spatial relationships of
the fingerprints as identified by human examiners, rather than lowlevel pixel values. These print comparisons were labeled similar
distractors, but it is important to note that the NAFIS algorithm is
a search aid and was not designed to model human performance;
so, what the NAFIS algorithm regards as similar may or may not
correspond to what a human examiner considers similar.
Establishing Expertise: Novices as a Control Group

Similarity
A pair of fingerprints will appear similar or dissimilar to each
other (or somewhere in-between), depending on the amount of
information in each and depending on the experience of the
examiner. There is no agreed upon definition or measure of similarity for the comparison of prints, but there have been attempts
to create an objective measure of similarity. For example, Vokey
et al. (15) converted a set of fingerprint images into their raw
pixel values (i.e., the brightness values in each fingerprint image)
and projected each print into the multidimensional space of all
the prints in a set to return a vector, where the similarity of one
print to another is given by the cosine of the angle between their
vectors. A cosine value close to 1 indicates that the prints are
virtually identical; whereas cosines close to zero indicate that the
prints are highly dissimilar. This technique, therefore, provides an
objective measure of similarity because it uses only the raw pixel
values in the images and so requires no human input. We did not
make use of this objective measure of accuracy for this experiment but, instead, used a national fingerprint database search.
For over 20 years, examiners have had the ability to search
large databases, with the aid of computer algorithms, for
potential matches to latent crime scene fingerprints. Although no
formal data exist, it is likely that the majority of fingerprint comparisons made today use database queries (suspect-absent cases)
rather than with a closed set of known prints from a suspect
(suspect-present cases). A database query on a latent print will
return a list of candidates that are most similar (according to the
algorithm) to that latent. As Vokey et al. (15) and Dror and
Mnookin (26) discuss, a database query makes an examiners
task much more difficult by returning a set of highly similar
distractor printsprints that look very much alike (according to
the algorithm) but come from different people. These searches,
by their very nature, are maximizing the conditions conducive to
false positives. Whats more, Vokey et al. (15) found that
novices made more false alarms on comparisons that were
similaras measured by the distance between vectors of pixel
mapsthan those that were not similar. Given that distinguishing highly similar, but nonmatching, prints from genuine prints
is likely to be the most difficult and common task that examiners
face, similarity was included as a factor in our experiment.
Similar distractor prints were obtained by searching our simulated crime scene latents on the NAFIS. The latents were first
auto-coded and then hand-coded by a qualified expert. For each
simulated crime scene print, the most highly ranked nonmatching
exemplar from the search was used if it was available in
the Queensland Police 10-print hard-copy archives, which
contains approximately one million 10-print cards (10 million
individual prints) from approximately 300,000400,000 people
(one person may have more than one 10-print card on record).
The corresponding 10-print card was retrieved from the archives,
scanned, and the individual print of interest was extracted. Due to

Research on the nature of human expertise has developed over

decades and ranges from the seemingly disparate domains of
chess to medical diagnosis. One goal of expertise research in
cognitive science is to understand the mechanisms that account
for the superior performance of experts, across domains (56).
Another is understanding the domain-specific nature of expertise:
Why does superior performance in one domain not transfer to
others? Can we hasten the transition from novice to expert?
Broadly, what makes an expert, an expert?
To study the nature of expertise in fingerprint identification, it
first needs to be demonstrated that expertise actually exists; that
is, are there people who possess exceptional abilities for matching
latent fingerprints to their source? Despite over 100 years of fingerprint testimonyand although implicit in the terminology
referring to fingerprint examiners who are qualified to testify in
court (i.e., a fingerprint expert)there had been no study demonstrating that qualified examiners have specialized discrimination skills or abilities superior to those of the person on the street
(but see [21] for a bias experiment with novices). Superior expert
discrimination performance in fingerprint identification had been
assumed, and the nature of that expertise had not been proposed
or demonstrated. But should people who have no training or experience be expected to accurately match prints to their source?
It is clear from Vokey et al. (15) that novices generally have
substantial abilities to match fingerprints. In a fingerprint matching task, nave undergraduates were able to discriminate fingerprint matches from nonmatches quite well, or well above chance
at least. With these findings in mindand without any requisite
experiments of expert performance in the forensic use of fingerprint identificationit was not obvious (to us at least) that
experts would outperform novices in our experiment as much as
they did. Furthermore, and as Vokey et al. (15) note, one pioneer of fingerprinting, Sir Francis Galton (57), believed that
experts would quickly become unnecessary and that lay juries
would eventually evaluate fingerprint evidence.
In 2002, Louis H. Pollak (a senior federal judge in Philadelphia; [58]) ruled, in United States v. Llera Plaza, that fingerprint
evidence does not meet the standards set for scientific testimony
and that experts in the field cannot testify that a suspects prints
definitely match those found at a crime scene. Pollak ruled that
fingerprint experts could still point out the similarities between
prints from a crime scene and those of a defendant, but the ultimate decision should be left to the jury. This decision was eventually overturned (59), but it is clear that one option for expert
testimony under consideration is for experts to present the physical evidence, with commentary attached, and allow lay juries to
decide whether a latent crime scene print matches the suspect.
Considering both the evidence for the reasonable performance
of novices and the notion that juries should make the ultimate
decision, it seems that the most appropriate comparison group to
demonstrate expertise should be novices who have no training

THOMPSON ET AL.

with fingerprints whatsoever. In the Identifying Fingerprint

Expertise experiment, the matching performance of qualified
fingerprint examiners was compared with the performance of
novice undergraduates who had no experience or training with
prints to establish the supposition of expertise in fingerprint
identification. Novices were 37 psychology undergraduates from
The University of Queensland who participated for course credit.
Experts were 37 qualified practicing fingerprint experts from five
police organizations (the Australian Federal, New South Wales,
Queensland, South Australia, and Victoria Police) who volunteered during our visit to their department. Their experience with
prints ranged from 5 to 32 years and was 17.45 (SD = 7.53)
years on average. We found that qualified, court-practicing fingerprint experts are exceedingly accurate compared with novices.
Even though the novices could reliably distinguish matching and
nonmatching prints, they made a large number of errors.
The performance difference between experts and novices on
trials in which the prints matched was relatively small (92.12%
correct for experts vs. 74.55% for novices). Comparably, the
performance difference between experts and novices on trials in
which the prints did not match, and were not similar, was also
relatively small (100% correct for experts vs. 77.03% for novices). The performance difference between experts and novices
on trials in which the prints were highly similar but did not
match, however, was substantial; novice participants mistakenly
identified 55.18% of the similar, nonmatching distractor prints as
matches, whereas the corresponding rate for experts was 0.68%.
The largest performance difference between novices and experts
seems to lie in identifying highly similar, but nonmatching
prints, as such. A comparison with novices was important for
demonstrating expertise and shows that the matching task
was difficult enough for experts to perform accurately, but for
novices to perform relatively poorly.
Error Rates
Much has been made about error rates in fingerprint identification, and more so in light of the NAS report (2). In this section, we attempt to characterize error rates in our experiment and
for fingerprint identification more generally.
Expert Matching Accuracy
In our experiment, 37 experts each compared 36 print pairs
for a total of 1332 comparisons. Of the 444 comparisons in
which the prints matched (targets), 22 of the 37 examiners incorrectly declared at least one of these matching prints as nonmatches, for an absolute total of 35 misses (hits = 92.12%;
misses = 7.88%). Misses are the kind of error that could allow a
guilty person to escape detection. Of the 444 comparisons in
which the prints did not match, and were not similar (nonsimilar
distractors), all of the examiners correctly declared the prints as
nonmatches (correct rejections = 100%; false alarms = 0%). Of
the 444 comparisons in which the prints did not match, but were
highly similar (similar distractors), three examiners incorrectly
declared three of these print pairs as matches (correct rejections = 99.32%; false alarms = 0.68%). These three print pairs
were made up of three different latents. False alarms are the kind
of error that could falsely incriminate an innocent person. (As an
aside, it is not possible to link a participants performance to the
identity of a particular individual because experts and novices
participated anonymously.) What, then, can be concluded about
error rates from this experiment?

FINGERPRINT EXPERTISE

1525

Determining Error Rates

Our study was not designed to determine the likelihood of
errors in practice, nor the performance of individual practitioners
or departments. It was designed to demonstrate expertise in
fingerprint identification. Inferring, from our results, that
Fingerprint examiners are 99.32% accurate, or The error rate
of fingerprint identification is 0.68%, would be unjustified. Any
claim of accuracy would have to be followed by a list of qualifiers. For example, Some qualified fingerprint examiners are
99.32% accurate at correctly declaring nonmatching prints as
such when the prints were obtained from the most similar
nonmatch according to the NAFIS, and when examiners are not
provided with their usual tools, or independent verification, and
so on. The qualifiers are limitations only when trying to make
an overgeneralization like, fingerprint examiners are 99.32%
accurate, which is close to impossible to make in any area of
expertise (let alone on the basis of our results). It is, however,
legitimate to conclude that experts are more accurate (and conservative) than novices, for example. If this is what we are concluding (and we are), then all of the qualifying remarks above
are completely irrelevant.
Readers might be now considering some particular qualifiers
to explain these results. For example, one might imagine that the
ability to zoom and rotate prints will improve expert performance or that verification will reduce experts error rate to zero.
But, it is unlikely that one particular qualifier will be enough to
entirely explain the results. Regardless, these qualifiers (and
many others) are all testable hypotheses, should the answers be
seen as necessary and important. If our goalin addressing the
critics and advancing the fieldis to determine the precise error
rate for each fingerprint examiner in each department, or even
the field as a whole, then the necessary experiments become
unwieldy.
We would need to unpack the different types of error that are
possible (e.g., clerical, identification, sufficiency, misses, false
alarms, disagreement, inadmissible rulings, etc.). We would need
to establish who is going to be tested (e.g., trainees, intermediates, qualified experts, supervisors), under what conditions (e.g.,
distractions, interruptions, sleep deprivation, time and resource
constraints, etc.), and on which part of the process (e.g., latent
development, analysis, comparison, testifying, etc.). Will we
include verification (and will it be blind)? How will we ensure
ground truth? What sort of materials will be examined (e.g., lifts,
surface types, lifting agents, powders, whorls, arches, low quality, distorted, highly similar, etc.)? What tools will be available
(e.g., image enhancements, digital markers, zoom, rotation, computer search algorithms, statistical models, etc.)? How much time
will they have? Can they collaborate? How many items will they
be tested on, and so on. In the end, we would be left with some
numbers (e.g., 85% hits and 6% false alarms for Jones under x,
y, and z conditions). What, then, do we do with this information? It is difficult to see how the incredible amount of time,
money, and resources required to get such an answer, would pay
off. As discussed earlier, this approach may be ineffective and
inefficient. And, we will be unable to locate the source of errors,
to say nothing of taking steps to avoid them.
Levels of Analysis
Is a measure of individual error necessary for the science or for
the court? The fingerprint profession, of course, will be concerned
with their performance to make sure that they are on track and to

1526

JOURNAL OF FORENSIC SCIENCES

ensure continuous improvement. And, of course, the courts will

be concerned with data that will help fact finders make optimal
decisions. But, for example, demands are not made for individual
error rates for a medical doctor or a field-wide error rate in medical diagnosis; only performance measures of the instrument or
test on average are sought. To ask for error rates associated with
a particular individual on a particular test seems, rightly, inappropriate in medicine. Similarly, focusing on the individual is the
wrong level of analysis when attempting to characterize the accuracy of forensic fingerprint identification processes and systems.
A broader question concerns the level of analysis that is
appropriate for presenting evidence and associated rates of error
in court. At the extreme, an examiner could report how accurate
they are at matching a whorl type print, lifted from a crime
scene, on a wooden surface, using magnetic black powder, in a
particular department, in a particular country, on a Tuesday, and
so on. It may be necessary, or the courts may demand, that particular rates of error are established for particular situations. But
unless it has been demonstrated that the level of accuracy (or
proficiency, or reliability, or competence) varies systematically
at any one of these levels, then the default should be to opt for
reporting accuracy at the broader level.
Error Rates in Other Domains
Comparing the matching performance of fingerprint experts
with experts in similar domains may give us an appreciation for
their relative performance. Unlike with fingerprints, all people
have expertise with faces. Psychologically, faces are similar to
prints in that they are both complex visual patterns (24). People
can easily recognize familiar faces despite changes in facial
expression, context, and viewpoint (60). Unfamiliar faces, however, are extremely difficult to identify across these changes.
Even in ideal conditions, where an unfamiliar target face is presented next to a set of candidate faceswith similar lighting,
poses and no time constraintspeople only match 68% of the
faces correctly (61). Even people whose job requires them to
identify unfamiliar faces from identity cards perform poorly at
this simultaneous matching task (62). Based on the results of
Tangen et al. (1), it is clear that fingerprint experts have impressive pattern matching abilities.
The accuracy of fingerprint experts becomes more impressive when compared with medical experts. Just as in fingerprint identification, however, it is difficult or impossible to
determine general rates of field-wide error. But it is known
that roughly 5% of autopsies reveal lethal diagnostic errors for
which a correct diagnosis coupled with treatment could have
averted death, and an estimated 40,00080,000 U.S. hospital
deaths result from misdiagnosis annually (63,64). These figures
suggest that more Americans are killed in U.S. hospitals every
6 months than died in the entire Vietnam War and is equivalent to three fully loaded jumbo jets crashing every other day
(but see [65]).
The prevalence of false positive diagnostic errors in perceptual
specialties, such as radiology and pathology, is typically <5%,
and increases to the range of 1015% in emergency room type
settings (66,67). Researchers are now focused on ways to reduce
(not eliminate) diagnostic errors and on creating policy that
defines acceptable rates of error (64). Of course it is difficult to
define error rates, let alone compare them across domains.
But, it is clear, from the available data, that fingerprint experts
demonstrate impressive pattern matching abilities that may rival
those of medical diagnosticians; even despite the distinction that,

arguably, identification (as in fingerprints) is a more difficult

task than categorization (as in medical diagnosis).
Summary
Thus far, we have expanded on the results of the Identifying
Fingerprint Expertise experiment (1) and explained that previous
experiments and tests of proficiency were problematic and that
the expertise of human fingerprint examiners had been assumed
but not demonstrated. We have described the decisions and compromises that we made to design an experiment that tests the
claimed expertise of human fingerprint examiners. In summary:
Fidelity, generalizability, and control must be balanced to
answer research questions. Our experiment was artificial
for good reason. The goal was to understand the extent to
which the difference between expert and novice performance
is real, not the extent to which the experimental setting
resembles the everyday operations of a fingerprint bureau.
The
validity, proficiency, and competence of fingerprint

examiners are best determined when experiments include

highly similar print pairs where the ground truth is known.
Prints from the Forensic Informatics Biometric Repository
were used to ensure ground truth.
To best quantitate matching performance, a signal detection
paradigm can be employed to separate the two ways of being
right and the two ways of being wrong, to compare performance on matching and nonmatching prints, and to separate
accuracy and response bias.
Distinguishing highly similar, but nonmatching, prints from
genuine prints is likely to be the most difficult and common
task that examiners face. Similar distractor prints were
obtained by searching simulated crime scene latents on the
NAFIS to emulate this task.
Considering both the evidence for the reasonable performance
of novices and the notion that juries should make the ultimate
decision, the most appropriate comparison group to demonstrate expertise should be novices who have no training with
fingerprints whatsoever.
Our study was not designed to determine the likelihood of
errors in practice, nor the performance of individual practitioners or departments. As such, inferring from our results
that, Fingerprint examiners are 99.32% accurate, or The
error rate of fingerprint identification is 0.68%, would be
unjustified.
Determining error rates with black box studies may be unnecessary at best and ineffective and inefficient at worst, and
unless one can demonstrate that a particular qualifier will
systematically affect accuracy, the default should be to report
accuracy at the broader level.
Fingerprint experts posses impressive pattern matching abilities that may rival those of medical diagnosticians.
It appears that expertise in fingerprint identification does exist.
That is, there are people who have demonstrable and specialized
abilities for matching latent fingerprints to their source, and
those abilities are superior to the person on the street. An examiners expertise seems to be situated, not in their ability to match
prints per se, but in their superior ability to identify highly similar, but nonmatching fingerprints as such. These results, and
their comparison with novices, show that the accuracy of qualified examiners is substantially higher than inexperienced novices. Moreover, the experiment was designed to be difficult. The
fact that experts made so few errors is evidence for impressive

THOMPSON ET AL.

human pattern matching performance possibly exceeding that of

experts in other comparable domains.
It seems that some combination of training and the daily comparison of untold numbers of fingerprints leads to an uncanny
ability to match fingerprints to their source. Experts are drawing
on an entire career of experiences in making their decisions, as
well as their training in fundamentals of fingerprint impressions
to understand the behavior of minutiae. Experts, likely implicitly, understand the structure, regularities, and acceptable variation
of fingerprint impressions. Future experiments could pinpoint the
nature of this expertise. That is, whether expertise arises mainly
from formal rules or the accumulation of instances (68).
Implications for Expert Testimony
The results from Tangen et al. (1) demonstrate that qualified
fingerprint experts perform much better than novices at matching
fingerprints, and their rates of error may be lower than those in
diagnostic medicine, for example. The implications of these
results for current models of expert admissibility, testimony, and
policy are discussed below.

FINGERPRINT EXPERTISE

1527

fingerprint evidencewhen presented with the physical evidence

alone, without expert commentary (56)could result in a substantial number of false identification errors.
The NAS Report (2,3) noted the frequent absence of solid scientific research demonstrating the validity of forensic methods in
general; of quantifiable measures of the reliability and accuracy
of forensic analyses; and of quantifiable measures of uncertainty
in the conclusions of forensic analyses. Fingerprint examiners
have taken a first step in demonstrating their claimed expertise
in controlled, representative situations in which ground truth is
known. Examiners are now working with researchers toward
understanding the source of identification errors, the factors that
influence performance, and the nature of expertise in identification. In light of the NAS report, and the model for demonstrating expertise provided here by fingerprint examiners and
researchers, it behooves other forensic pattern identification
disciplinessuch as shoeprints, bloodstains, DNA, ballistics,
toolmarks, bitemarks, CCTV face identification, etc.to conduct
similar experiments to demonstrate expertise and performance in
their own disciplines.
Testimony

The Current Model

Current models of expert testimony vary from country to
country and from state to state. The two largest bodies that provide consensus guidelines and standards for fingerprint identificationthe Scientific Working Group on Friction Ridge
Analysis, Science and Technology (69) and the International
Association for Identification (70)both stipulate that examiners
are only permitted to testify to three conclusions: exclusion,
inconclusive, and individualization. An individualization is
defined as, The determination by an examiner that there is sufficient quality and quantity of detail in agreement to conclude
that two friction ridge impressions originated from the same
source (71, p. 6). When testifying, examiners often do not provide evidence of their claimed expertise or attempt to characterize their level of proficiency. Examiners may, when pushed by
the courts, report that all fingerprints are unique or point to
prenatal development and persistence. Despite considerable acrimony (2), examiners continue to make claims of individualization or similar ([72], but see [11]).

The Implications of the Identifying Fingerprint Expertise

Experiment
Admissibility
Information about accuracy and performancealong with the
relative performance of laypersonsis required for courts to
make informed decisions about the admissibility of expert testimony. Experts outperform novices, but they do make errors (1).
These results make it less likely that examiners themselves will
suffer unfounded attacks on their expertise. If an examiners
expertise is challenged, then our methodology and design
ought to be the target of criticism rather than the examiners
themselves; assuming, of course, that their testimony does not
extend beyond what our experiments can support.
The distinction, between the performance of experts and novices, is fundamental to the question of expert testimony, because
it demonstrates specialized knowledge. Our experiment could be
used as evidence for this distinction to satisfy legal admissibility
criteria. And the results suggest that relying on juries to evaluate

Documented cases of false identification (9), issues of plausibility reported by the NAS (2), and recent experiments (1,32
34) highlight the need for a contemporary model of forensic
testimony. Following developments in the United States and Canada, Edmond (73) has suggested that Australia adopt a reliability
standard, and the U.K. Law Commission (4) has announced similar recommendations for admissibility practice in England and
Wales. Indeed, science and legal commentators are beginning to
call for empirical demonstrations of accuracy and performance,
along with details about the relative performance of laypersons,
across forensic science. A failure to respond to criticism means
that judges are in danger of acting irrationally and being left
behind by practical and ongoing reforms in the forensic sciences.
While it is likely that courts will start to develop an admissibility
and testimony jurisprudence more directly concerned with reliability in the near future, there is an independent need for forensic
scientists and technicians to pay much closer attention to the evidence for ability and reliability (74). Edmond (73) suggests that
reliability standards will help to make criminal trials fairer and
ensure outcomes reflect the known value of expert evidence.
It is clear that an alternative to the current model of fingerprint testimony is required. But, what should an acceptable alternative and contemporary model look like? Several factors must
be considered; these include the role of scientific experiments on
the accuracy and reliability of forensic identification; whether it
is necessary to report department or individual scores on properly controlled proficiency tests; the state of the science in other
areas of pattern and impression identification; the impact that the
testimony has on jury decision making; finding the right balance
between accurate scientific reporting and the ability of judges
and juries to understand expert testimony; and decisions about
whether to report on the degree to which the specimen matches
the source (e.g., lends limited support), the degree of confidence in a match (e.g., highly confident that x matches y),
opinions about the evidence (e.g., it is my opinion that), or
statements about the particular hypotheses in question (e.g., the
evidence is more consistent with x than y).
There is much research and consideration needed to develop an
acceptable alternative model of fingerprint testimony. Several
debates on this topic are raging internationally between academics,

1528

JOURNAL OF FORENSIC SCIENCES

statisticians, lawyers, forensic examiners, and managers. We are

working toward proposing recommendations that do not extend
beyond the capabilities of examiners or experimental findings
while substantially engaging with critics to develop robust
empirical guides to practice.
To Develop a Research Culture in Forensic Science
Researchers and professionals (e.g., [40]) have highlighted the
need for a research culture in forensic science. Currently, however, it appears that professionals are expected to strengthen the
scientific basis of their field but are not provided with the financial or intellectual support to do so. It is clear that examiners are
doing their best to capture criminals and uphold civil liberties.
But the lack of funding and resources in already overworked
forensic departments makes basic research exceedingly unlikely.
In addition, few have the methodological skills and expertise in
the psychology of perception, cognition, bias, memory, accuracy,
and decision making to ensure that their practice meets legal
admissibility standards emerging internationally.
It is essential that we move beyond the adversarial system currently impeding advancement of the field and develop a culture
of cooperation between researchers and examiners. The emergence of such a culture would fundamentally change examiners
relationship with empirical data and affect how evidence is
understood and reported. Indeed, forensic examiners have
expressed a desire to address the shortfalls of their discipline
and engage in research.
Considering that forensic identification is based on human
judgment, the field would benefit from further research on expert
decision making. Clinical reasoning in medicine, for example,
has developed over the last 40 years after it became increasingly
apparent that physicians decisions resulted in adverse consequences for patients (63). Much has been learned about the nature of medical expertise, the influence of perceptual and
cognitive biases, and how to best incorporate such knowledge
into practice. Researchers need to provide a scientific basis for
demonstrating the validity of forensic methods and measures of
uncertainty in the judgments of forensic analyses.
From here, more sophisticated questions can be asked than
those about error rates and expertise. For example, what is the
most effective way to train novices? What information is the
most important for matching or excluding prints? What elements
of the matching task best distinguish experts and novices? How
do experts and novices differ in their use of this information?
How does expertise with fingerprints develop over time? What
is the relationship between the Analysis and Comparison phase
of the identification process? How does time pressure influence
performance? What is best practice in providing feedback and
self-assessment? What is the most effective way to present fingerprint evidence to juries? The practical outcomes from answering questions such as these include a better understanding of the
source of potential identification errors and factors that influence
performance, a reduction in training time from novice to expert,
more effective recruitment and training methods, and greater
validity in presenting forensic evidence in court.
Maintaining high standards of evidence is vital for an effective
justice system and ensuring that innocent people are not wrongfully accused. The reliability of forensic evidence and the value
of expert testimony in the criminal justice system can be maximized by examining forensic reasoning and decision making.
Given the inevitability of human error, the move should be
toward fostering resilient systems capable of minimizing and

acknowledging errors (75). Wecollectively, forensic professionals, researchers, legal scholars, and the courtsneed to
define acceptable rates of error, foster a work environment
conducive to learning from error, and promote a blame-free
safety culture, as medicine is working toward (64,76).
This approach will allow police, intelligence systems, and
investigators to interpret evidence more effectively and efficiently, assist forensic examiners in the development of evidencebased training programs, discourage exaggerated interpretations
of forensic evidence, and help in the development of a model of
expert testimony that does not extend beyond the capabilities of
examiners or beyond the scope of experimental findings. Further
psychological research into forensic decision making will help to
ensure the integrity of forensics as an investigative tool available
to police, so the rule of law is justly applied.
Acknowledgments
We thank Professor Gary Edmond for his valuable input
toward this publication and Morgan Tear and Hayley Thomason
for assistance with data collection.
References
1. Tangen JM, Thompson MB, McCarthy DJ. Identifying fingerprint expertise. Psychol Sci 2011;22(8):9957.
2. National Research Council, Committee on Identifying the Needs of the
Forensic Science Community. Strengthening forensic science in the United States: a path forward. Washington, DC: The National Academies
Press, 2009.
3. Edwards HT. Statement of The Honorable Harry T. Edwards: strengthening forensic science in the United States: a path forward. Washington,
DC: United States Senate Committee on the Judiciary, 2009.
4. Law Commission of England and Wales. Expert evidence in criminal
proceedings in England and Wales. London: Stationery Office, 2011.
5. Loftus EF, Cole SA. Contaminated evidence. Science 2004;304
(5673):959.
6. Saks MJ, Koehler JJ. The coming paradigm shift in forensic identification science. Science 2005;309(5736):8925.
7. Spinney L. The fine print. Nature 2010;464:3446.
8. Federal Bureau of Investigation. The science of fingerprints: classification and uses. Washington, DC: U.S. Government Printing Office, 1984.
9. Cole SA. More than zero: accounting for error in latent fingerprint identification. J Crim Law Criminol 2005;95(3):9851078.
10. Cole SA. Who speaks for science? A response to the National Academy
of Sciences report on forensic science. Law Prob Ris 2010;9:2546.
11. Garrett R. Memorandum from the President of the International Association for Identification, February 19, 2009, http://www.theiai.org/current_
affairs/nas_memo_20090219.pdf. (accessed December 1, 2011).
12. Scientific Working Group on Friction Ridge Analysis Study and Technology (SWGFAST). Standard for the definition and measurement of
rates of errors and non-consensus decisions in friction ridge examination
(latent/tenprint). Ver. 1.1, 09/16/11, SWGFAST, 2011.
13. Cole SA, Welling M, Dioso-Villa R, Carpenter R. Beyond the individuality of fingerprints: a measure of simulated computer latent print source
attribution accuracy. Law Prob Ris 2008;7(3):16589.
14. Haber L, Haber RN. Scientific validation of fingerprint evidence under
Daubert. Law Prob Ris 2008;7(2):11926.
15. Vokey JR, Tangen JM, Cole SA. On the preliminary psychophysics of
fingerprint identification. Q J Exp Psychol (Hove) 2009;62(5):102340.
16. Dror I, Charlton D. Why experts make errors. J Forensic Identif 2006;56
(4):60056.
17. Dror I, Charlton D, Peron A. Contextual information renders experts vulnerable to making erroneous identifications. Forensic Sci Int 2006;156
(1):748.
18. Dror I, Cole S. The vision in blind justice: expert perception, judgment, and visual cognition in forensic pattern recognition. Psychon B
Rev 2010;17(2):1617.
19. Dror I, Peron A, Hind S, Charlton D. When emotions get the better of
us: the effect of contextual top-down processing on matching fingerprints. Appl Cognitive Psychol 2005;19(6):799809.

THOMPSON ET AL.
20. Dror I, Rosenthal R. Meta-analytically quantifying the reliability and
biasability of forensic experts. J Forensic Sci 2008;53(4):9003.
21. Langenburg G, Champod C, Wertheim P. Testing for potential contextual
bias effects during the verification stage of the ACE-V methodology when
conducting fingerprint comparisons. J Forensic Sci 2009;54(3):57182.
22. Busey T, Dror I. Special abilities and vulnerabilities in forensic expertise.
In: Holder EH Jr, Robinson LO, Laub JH, editors. Fingerprint sourcebook. Washington, DC: National Institute of Justice Press, 2010;15-1
15-23.
23. Busey T, Parada F. The nature of expertise in fingerprint examiners.
Psychon B Rev 2010;17(2):15560.
24. Busey T, Vanderkolk J. Behavioral and electrophysiological evidence for
configural processing in fingerprint experts. Vision Res 2005;45(4):
43148.
25. Busey T, Yu C, Wyatte D, Vanderkolk J, Parada F, Akavipat R. Consistency and variability among latent print examiners as revealed by eye
tracking methodologies. J Forensic Identif 2011;60(1):6191.
26. Dror IE, Mnookin JL. The use of technology in human expert
domains: challenges and risks arising from the use of automated fingerprint identification systems in forensic science. Law Prob Ris
2010;9(1):4767.
27. Dror IE, Wertheim K, Fraser-Mackenzie P, Walajtys J. The impact of
human-technology cooperation and distributed cognition in forensic
science: biasing effects of AFIS contextual information on human
experts. J Forensic Sci 2012;57(2):34352.
28. Champod C, Evett I. A probabilistic approach to fingerprint evidence.
J Forensic Identif 2001;51(2):10122.
29. Neumann C, Champod C, Puch-Solis R, Egli N, Anthonioz A, BromageGriffiths A. Computation of likelihood ratios in fingerprint identification
for configurations of any number of minutiae. J Forensic Sci
2007;52:5464.
30. Neumann C, Champod C, Puch-Solis R, Egli N, Anthonioz A, Meuwly
D, et al. Computation of likelihood ratios in fingerprint identification
for configurations of three minutiae. J Forensic Sci 2006;51(6):1255
66.
31. Neumann C. Fingerprints at the crime-scene: statistically certain, or probable? Significance 2012;9(1):215.
32. Ulery BT, Hicklin RA, Buscaglia J, Roberts MA. Accuracy and reliability of forensic latent fingerprint decisions. Proc Natl Acad Sci USA
2011;108(19):77338.
33. Ulery BT, Hicklin RA, Buscaglia J, Roberts MA. Repeatability and
reproducibility of decisions by latent fingerprint examiners. PLoS ONE
2012;7(3):e32800.
34. Dror IE, Champod C, Langenburg G, Charlton D, Hunt H, Rosenthal R.
Cognitive issues in fingerprint analysis: inter- and intra-expert consistency and the effect of a target comparison. Forensic Sci Int 2011;208
(1):1017.
35. Langenberg G. Performance study of the ACE-V process: a pilot study
to measure the accuracy, precision, reproducibility, repeatability, and
biasability of conclusions resulting from the ACE-V process. J Forensic
Identif 2009;59(2):21957.
36. Wertheim K, Langenburg G, Moenssens A. A report of latent print examiner
accuracy during comparison training exercises. J Forensic Identif 2006;56
(4):5593.
37. Haber L, Haber RN. Letter to the editor. Re: a report of latent print examiner
accuracy during comparison training exercises. J Forensic Identif 2006;56
(4):4939.
38. Wertheim K, Langenburg G, Moenssens A. Authors response to letter:
letter to the editor. Re: a report of latent print examiner accuracy during
training exercises. J Forensic Identif 2006;56(4):50010.
39. FBI responds to the office of inspector generals report on the fingerprint
misidentification of Brandon Mayfield. 2006, http://www.fbi.gov/news/
pressrel/press-releases/fbi-respondsto-the-office-of-inspector-general2019sreporton-the-fingerprint-misidentification-of-brandon-mayfield (accessed
November 3, 2011).
40. Mnookin JL, Cole SA, Dror IE, Fisher BAJ, Houck M, Inman K, et al.
The need for a research culture in the forensic sciences. UCLA Law Rev
2011;58(3):72579.
41. Spinney L. Forensic science braces for change. Nature 2010, doi:
10.1038/news.2010.369, http://www.nature.com/news/2010/100722/full/news.
2010.369.html (accessed December 1, 2011).
42. Dror IE. The paradox of human expertise: why experts get it wrong. In:
Kapur N, editor. The paradoxical brain. Cambridge: Cambridge University Press, 2011;17788.
43. Dror IE. Combating bias: the next step in fighting cognitive and psychological contamination. J Forensic Sci 2012;57:2767.

FINGERPRINT EXPERTISE

1529

44. Koehler JJ. Fingerprint error rates and proficiency tests: what they are
and why they matter. Hastings Law J 2007;59:10771110.
45. Thompson MB, Tangen JM, McCarthy DJ. Human matching performance of genuine crime scene latent fingerprints. Law and Hum Behav
2013 In press.
46. Mook DG. In defense of external invalidity. Am Psychol 1983;38
(4):37987.
47. Brinberg D, McGrath JE. Validity and the research process. Beverly
Hills, CA: SAGE Publishing Co, 1985.
48. Woods DD. The observation problem in psychology (Westinghouse
Technical Report). Pittsburgh, PA:Westinghouse Corporation, 1985.
49. Brunswik E. Perception and the representative design of psychological
experiments, 2nd rev. ed. & enl. Berkeley, CA: University of California
Press, 1956.
50. Rasmussen J, Pejtersen AM, Goodstein LP. Cognitive systems engineering. New York, NY: John Wiley & Sons, Inc, 1994.
51. Sanderson P. Capturing the essentials: simulator-based research in aviation and healthcare. Proceedings of the Eighth International Symposium
of the Australian Aviation Psychology Association; 2008 Apr 811;
Sydney, Australia. Sydney: Australian Aviation Psychology Association,
2008.
52. Sanderson P, Liu D, Jenkins S, Watson M, Russell WJ. Summative display evaluation with advanced patient simulation: fidelity, control, and
generalizability. Brisbane, QLD: Cognitive Engineering Research Group,
The University of Queensland, 2010. Report No.: CERG-2010-01.
53. Haber L, Haber RN. Error rates for human fingerprint examiners. In:
Rath NK, Bolle RM, editors. Automatic fingerprint recognition systems.
New York, NY: Springer-Verlag, 2004;33960.
54. Phillips VL, Saks MJ, Peterson JL. The application of signal detection
theory to decision-making in forensic science. J Forensic Sci
2001;46:294308.
55. Green DM, Swets JA. Signal detection theory and psychophysics. New
York, NY: Wiley, 1966.
56. Ericsson KA, Smith J. Toward a general theory of expertise: prospects
and limits. Cambridge: Cambridge University Press, 1991.
57. Galton F. Decipherment of blurred fingerprints: supplementary chapter to
finger prints. London: Macmillan, 1893.
58. Cho A. Fingerprinting doesnt hold up as a science in court. Science
2002;295(5554):418.
59. Cho A. Judge reverses decision on fingerprint evidence. Science
2002;295(5563):21957.
60. Johnston R, Edmonds A. Familiar and unfamiliar face recognition: a
review. Memory 2009;17(5):57796.
61. Megreya AM, Burton AM. Matching faces to photographs: poor performance in eyewitness memory (without the memory). J Exp Psychol-Appl
2008;14(4):36472.
62. Kemp R, Towell N, Pike G. When seeing should not be believing: photographs, credit cards and fraud. Appl Cognitive Psychol 1997;11
(3):21122.
63. Institute of Medicine. To err is human: building a safer health system.
Washington, DC: National Academy Press, 2000.
64. Newman-Toker DE, Pronovost PJ. Diagnostic errorsthe next frontier
for patient safety. JAMA 2009;301(10):10602.
65. Hayward RA, Hofer TP. Estimating hospital deaths due to medical
errors. JAMA 2001;286(4):41520.
66. Berner ES, Graber ML. Overconfidence as a cause of diagnostic error in
medicine. Am J Med 2008;121(5):S23.
67. Norman GR, Eva KW. Diagnostic error and clinical reasoning. Med
Educ 2010;44(1):94100.
68. Norman GR, Young M, Brooks L. Non-analytical models of clinical reasoning: the role of experience. Med Educ 2007;41(12):11405.
69. Scientific Working Group on Friction Ridge Analysis Study and Technology (SWGFAST). Standards for Conclusions. Ver. 1.0, 9/11/03,
SWGFAST, 2003.
70. International Association For Identification (IAI). IAI position concerning
latent fingerprint identification, 2007, www.onin.com/fp/IAI_Position_
Statement_11-29-07.pdf. (accessed December 1, 2011).
71. Scientific Working Group on Friction Ridge Analysis Study and Technology (SWGFAST). Standard terminology of friction ridge examination.
Ver. 3, 3/23/11, SWGFAST, 2011.
72. Cole SA. Forensics without uniqueness, conclusions without individualization: the new epistemology of forensic identification. Law Prob Ris
2009;8(3):23355.
73. Edmond G. Specialised knowledge, the exclusionary discretions and
reliability: reassessing incriminating expert opinion evidence. UNSW L
J 2008;31(1):155.

1530

JOURNAL OF FORENSIC SCIENCES

74. Edmond G. Actual innocents? Legal limitations and their implications

for forensic science and medicine. Aust J Forensic Sci 2011;43(2):177
212.
75. Hollnagel E, Woods DD, Leveson N. Resilience engineering: concepts
and precepts. Alershot: Ashgate Publishing Company, 2006.
76. Woods DD, Johannesen L, Dekker S, Cook R, Sarter N. Behind human
error, 2nd edn. Burlington, VT: Ashgate Publishing, Ltd., 2010.

Additional information and reprint requests:

Matthew B. Thompson, B.Inf.Tech., B.Sc.
School of Psychology
The University of Queensland
St Lucia QLD 4072
Australia
E-mail: [email protected]