Evidence for the Memory Color Effect 1

Tutorial Review: Evidence for the Memory Color Effect, 1923–2016

Ifedayo Emmanuel. Adeyefa-Olasupo ^*#! & Jonathan I. Flombaum^

^Department of Psychological and Brain Sciences, Johns Hopkins University,

*Berlin School of Mind and Brain, Humboldt-Universität zu Berlin
# Department of Psychiatry and Psychotherapy, Charité Universitätsmedizin Berlin,
! Department of Neuroscience, Yale School of Medicine

Address for Correspondence:

Ifedayo-Emmanuel Adeyefa-Olasupo
Email: [email protected]
Mailing address:
200 S Frontage Rd
Yale School of Medicine
New Haven, CT 06511

Running Head: Evidence for the Memory Color Effect

Key words: Perception, Memory, Cognitive Penetrability, Top-Down, Encapsulation
Word count:7952 Abstract; 3 Figures.
Evidence for the Memory Color Effect 2

Abstract
Is the appearance of an object solely determined by incoming sensory data? If not, to what extent

is this sensation calibrated by what is memorized or know about the object? Under normal

condition, do bananas, for example, appear more yellow because people know and or remember

that bananas are yellow (most of the time), compared to other objects (all lighting and reflectance

properties held constant)? Empirical research on this possibility has a long history, and it enjoys

the reputation of perhaps the best example of an effect of cognition on perception. Here we review

studies that are frequently cited as positive evidence, beginning in 1923 and continuing until

present time. The intent is to provide a general sense of the methods that have been applied to the

problem, and to conclude with a qualitative impression of the strength of extant evidence. Towards

this end, we identify a number of pitfalls that persistently complicate the interpretation of results.

And we highlight what we take to be the best evidence currently available, with suggestions for

extending that work.

Evidence for the Memory Color Effect 3

General Introduction

What governs the visual appearances of objects? An object’s reflectance properties, the

external lighting conditions, and an observer’s relative viewing orientation should all matter, along

with any other factors that impact the specific properties of received sensory data. But these data

are known to be insufficient with respect to the inverse problem of perception. The fixing of

appearances must therefore involve inference, aided by some forms of prior knowledge (Marr

1982; Kanizsa 1985; Knill & Richards, 1996). So much is consensus. Vigorously debated are the

kinds of prior knowledge that influence perception, where and when in processing. Of particular

interest is whether explicit knowledge, intentions, and beliefs can affect perception, and

specifically in terms of what things look like to the first-person observer. We will use the term

‘appearance’ to refer to how things look to an observer in order to avoid confusion with the

potentially broader implications of the term ‘perception.’ Intuitively, it seems reasonable that

explicit knowledge should support the fixing of objects’ appearances. If one knows that bananas

are usually yellow, for example, then shouldn’t this knowledge alleviate some of the ambiguity

inherent in disentangling surface color from lighting properties? Perhaps; but there are countless

and equally intuitive places where explicit knowledge patently fails to have an impact. Consider

virtually any visual illusion that would be shared during a standard lecture on sensation and

perception. The illusion can be familiar, and its causes are likely understood in detail. Yet the

stimulus’ illusory appearance persists. The well-known Muller-Lyer illusion is a simple example.

Understanding exactly why the illusion obtains does not prevent it.

There is a great deal at stake theoretically with respect to whether explicit knowledge

affects appearance, even if only a little and only sometimes. That literature is too large to review
Evidence for the Memory Color Effect 4

here. So we point interested readers to several extended discussions and debates, where effects of

explicit knowledge on perception are often referred to as demonstrating the cognitive penetrability

of perception. Useful references include: Firestone & Scholl (2015), Machery (2015), Firestone

(2013b), Lupyan (2012), Proffitt & Linkenauger (2013), Goldstone & Barsalou (1998), Pylyshyn

(1999), (2002), and Fodor, (1988), (1983), (1984).

Here we closely consider reports in the domain of one especially important case: the case

of canonical color knowledge affecting color appearance, sometimes called ‘The Memory Color

Effect’ (MCE). An example of an MCE would be any case like that of bananas described already.

Does knowing that certain objects tend to possess certain colors affect the color appearances of

those objects under normal conditions?

This effect is of particular importance for at least two reasons. One is simply that it has a

good reputation as a positive effect, with a number of high profile reports relatively recently (e.g.

Hansen et al. 2006; Olkkonen et al. 2008; Mitterer et al. 2008 2009; Witzel et al. 2011; Kimura et

al.2013; Witzel 2016) More importantly for current purposes, it is a case where the claim is

circumscribed to color perception, as opposed to being more broadly framed in terms of the

flexibility of perception. The MCE therefore has specific implications for how color perception

may work, an area with rigorous and well-established models. If memory really affects color

appearance, then those models require a channel by which the effects arise. In what follows we

review what perceive as the most highly-cited, compelling, or otherwise representative and

instructive studies of the MCE in peer-reviewed outlets since 1923. All the papers that we are

aware of are cited here, although we do not seek for our discussions to be exhaustive. Instead, we

review papers with three goals in mind: (1) to comprehensively identify the main methods that

have been used in this research area. (2) To identify the small number of recurring challenges that
Evidence for the Memory Color Effect 5

complicate the interpretation of published results. (3) To emphasize the most methodologically

rigorous approaches and the most compelling results to date. We conclude with our own

impression that the available evidence is weaker than a casual observer may have supposed, and

with some suggestions for what stronger evidence might look like.

Studies 1923-1965

For reasons having to do with changes in methodology we will mark the first phase of experiments

on the MCE as ranging from 1923-1965.

Adams 1923. The earliest modern study to look for a memory colour effect focused on the

question of whether observers could learn a color-object association, rather than focusing on

pervasive, every-day associations. The study involved two phases. In the training phase, a subject

was exposed (over 1000 times) to a blue jar (an object which does not have a typical color),

fabricated from blue construction paper, and shown under normal lighting conditions. The subjects

reported the color of the jar using a color wheel. In the second phase of the experiment, the test

phase, subjects were presented with the same jar, this time fabricated from grey construction paper,

and presented in dim lighting (low luminance). There were asked to now report the color of the

gray urn, and 83% of subjects adjusted the urn with a bias in the blue direction —the colour that

the vase was associated with through exposure in part one of the experiment. Methodologically,

the study employs an intuitive approach, which we will call the method of Direct Report:

participants were instructed to report directly the color that they saw at the time of test. This is in

contrast to more contemporary methods that ask participants to make adjustments that reflect what

they see but are not framed as direct reports of what they see, for example, methods that ask

participants to adjust a stimulus to match a background, or to achieve an achromatic appearance.

Evidence for the Memory Color Effect 6

The simplest alternative account of these results is in terms of subject accommodation and

experimental transparency. With only a single condition, and fewer than 10 participants, the

transparent nature of the expected results complicates any interpretation. With respect to other

forms of knowledge affecting perception, recent experiments have demonstrated how experimental

transparency leads frequently to accommodation (Durgin et al. 2009).

A more technical problem has to do with the specific colors used. Winker and colleagues

(2015) demonstrated that when subjects are presented with a homogenously grey stimulus, subtle

chromatic variation along the blue axis which is perceived as the illuminant causes subjects to

perceive the stimulus as being closer to their own achromatic locus. The coincidental use of grey

and blue as the relevant colors in this experiment are therefore unfortunate, in retrospect, making

the results hard to interpret. In general, effects from a single color category should be considered

weak evidence at best, and replication with additional colors should be demanded before

conclusions drawn.

Interestingly, the paper describes experiments on canonical odor and texture, while also

reviewing in detail Herring’s experiments on memory color (which involved making a small

incision in the skin of an assistant in order to investigate the perception of blood). It is well worth

reading, and it appropriately invited further investigation of the memory color effect. Duncker

1939. Subjects were presented with a high colour diagnostic stimulus (green artificial leaves) and

a controlled stimulus (a green mule cutout). One at a time, the stimuli were placed in a chamber

with a plate of red glass through which the observer could see the stimuli. The idea was that under

red lighting —the opponent hue— a green stimulus should appear achromatic (e.g. Hurvich &

Jameson 1957). In an adjacent chamber was a color wheel under normal lighting. Participants

looked back and forth between the wheel and test stimuli, instructing the experimenter in which
Evidence for the Memory Color Effect 7

direction to turn the wheel in order to match the color of the test stimulus. The prediction was that

the achromatic appearance of the leaf would be greener than the mule, i.e. that the reported ‘gray’

of the leaf should be biased in the green direction.

The exact quantitative results are difficult to discern because the methods and the

presentation are not specific enough (e.g. “When, after about 6 trials, the leaf equation had been

obtained, the leaf was again exchanged for the donkey…” pp. 260). Moreover, Duncker reports

the results of only six of the 11 subjects, specifically those who showed effects. He argues that the

other five did not show effects because of lack of interest in the experiment. By modern standards,

the data and methods fall short. Nonetheless, the study has been cited at least three dozen times

since 2000, as positive evidence and without caveats. This is problematic. We cite the paper here,

in part, to emphasize that converging evidence can appear stronger than it should when historical

studies are not reevaluated.

The study is also illustrative of a two methodological features of early work in the area.

First, like study by Adams (1923) the mode of response is a direct report. Second, in this study,

the participants delivered oral instructions concerning how the color wheel should be manipulated

to triangulate a final answer. The instructions were then implemented by (an unblinded) researcher,

something like a game of ‘hot and cold.’ Thus, final responses were experimenter intermediated.

This approach was common, although it obviously also falls short by contemporary standards.

Bruner et al 1951. Observers were presented with several high colour diagnostic stimuli

(lemon, banana, tomato, tangerine, boiled lobster claw, carrot) fabricated in gray and placed

against a bluish-green background. This was done to induce a chromatic contrast effect (Figure 2),

wherein an achromatic stimulus appears desaturated owing to contrast with a bluish-green

Evidence for the Memory Color Effect 8

background. The observers were split between two conditions. Group 2 observers were familiar

with the expected chromatic contrast effects while Group 1 observers were not. The prediction was

that group one observers would show a memory color effect: that they would report the color of a

grey canonical stimulus as biased in the direction of its canonical color. The prediction for group

2 was they would show less of an effect —not because they did not experience a MCE, but because

their knowledge of the contrast effect would produce a countervailing expectation. Bruner

reasoned that if knowledge of canonical color can have an effect on appearance, then so should

other knowledge, in this case, knowledge of a contrast effect. This prediction did not manifest.

Instead, both groups showed memory color effects of equal magnitude.

The study also included another two conditions. One group of participants saw canonical

objects that were recognizable but varied away from their canonical shapes (e.g. a straighter than-

usual banana). The other group saw canonically-shaped canonical objects. The prediction here was

that the first group would show weaker effects, that the typicality of an objects shape should

continuously modulate the strength of a memory color effect. This prediction also did not come to

pass. Both groups showed equal effects. What to make of these results? Bruner’s experiments, had

they worked as predicted, would have supplied strong evidence of a MCE. The attempt was to

move beyond a binary present/absent set of results to more continuous measures and predictions.

But because the results remained binary, the experiments suffer from the confound of subject

accommodation. That is, the results may not reflect that the objects appeared canonically colored

to subjects, and instead are open to the interpretation that they only reported them as such, knowing

what was expected in the experiment. Indeed, Bruner’s New Look research program has been

generally criticized on this basis (see e.g. Bruner & Postman, 1948 and Klein and colleagues,

1951). Nonetheless, in the case of color, Bruner’s work remains highly cited as a positive evidence.
Evidence for the Memory Color Effect 9

Bruner’s failed predictions point also to subtler signs of false-positive effects. Bruner predicted

that one form of prior knowledge —familiarity with the contrast effect— should countervail the

memory effect. But only the memory effect appeared. On an ad hoc basis, such a failed prediction

is easy to get around: the MCE must be a dominant type of prior knowledge. But this convenient

explanation neglects to worry about the presence of an effect where one should not be. Firestone

and Scholl (2013b) have emphasized in other contexts that showing affects where they should not

appear is a straightforward way to identify effects that are easily prone to false positives, whether

by accommodation or otherwise. The absence of a predicted effects should be worrisome here as

well.

The failed prediction concerning the continuity of the effects is worrisome as well, in this

regard. But more broadly, it highlights a problem that we believe to be pervasive, even in the best

contemporary experiments: A failure of continuous measure in the case of an effect that is

inherently continuous. In all cases, and regardless of the exact methods employed, the difference

in appearance between the color of a canonical and a non-canonical object, all else equal, is a

difference should be, by all accounts, continuously quantifiable. This follows from the mere nature

of color, the mere fact that something cannot be more yellow without being a specific yellow, a

specific color, that is, that could in principle be designated by three values in a color space. Yet

not Bruner, nor any study that we are aware of, has succeeded in leveraging the inherently

continuous nature of color to show a continuous effect. We can only say that in some experiments

bananas (as an example) appear significantly more yellow than non-bananas (all else equal). How

much more however seems to be an arbitrary and random variable that varies inexplicably across

participants and experiments, not systematically according to any obvious stimulus properties. In
Evidence for the Memory Color Effect 10

the case of color, the lack of continuous effects after nearly 100 years of experiments strikes us as

cause for concern.

Figure 1. The chromatic contrast effect. A. - D. As the chromaticity of the surround is shifted along

the bluish-green axis, the appearance of the center stimuli appears to be tinted towards an orangish-

brown hue. The perceptual consequence of this is that the center stimuli should appear desaturated,

particularly in the case of D.

Harper 1953. This study is oddly cited as supplying positive evidence of a memory color

effect. But the results of the study as a whole were inconsistent. Harper investigated only

canonically red objects: a lobster claw, a heart, and an apple (thereby suffering from the single

color category). As a control stimulus, he used the letter ‘Y.’ Cutouts of these objects were placed

on a neutral background, and participants instructed the experimenter to adjust a colorimeter to

cause the neutral background to match the hue of the cutout (The prediction was that the adjustment

would be to a hue that is redder than the hue of the cutouts, since the red cutouts themselves should

appear redder to the participants.

Methodologically, this is another study where accommodation effects should be expected

(the purpose of the study is transparent), and where a biased and un-blinded experimenter

participated directly in the adjustments on the basis of the participant’s oral instructions

(experimenter intermediated responding). The results are also problematic because the red heart

did not produce the desired effect (the other canonical stimuli did). And the Y control stimulus did
Evidence for the Memory Color Effect 11

produce the effect, producing the presence of an effect where one should not be. Yet the

experiment seems to be cited as uncomplicated positive support for the MCE (see also Heurley et

al. 2012 p. 365).

Fisher et al. 1956 . This is a rare published study that actually questioned the validity of

the extant research to this point. Fisher and colleagues were concerned that previous experiments,

particularly those of Bruner and Harper, were conducted without proper control of ambient

illumination. The experiment therefore sought to replicate the experiments of Harper (1953), with

the illuminant rigorously controlled across the different trials and conditions. Small and

insignificant effects were observed for some stimuli (e.g. a red cross in Experiment 2) and no effect

was found for another set of stimuli (e.g. a lobster claw, heart, and apple in Experiment 1). Fisher

et al. conclude that the memory colour effect is minimal and virtually non-existent. This is a rare

case where a null result that would normally end up in a file drawer was published, seemingly

owing to the emphasis placed on illumination control. The study, unfortunately, is cited less

frequently than the others already mentioned, indicating that the literature has failed to reconcile

conflicting results and failed replications.

Newhall et al. 1957. This study sought to examine the possibility that semantic activation

along with canonical shape should produce a stronger MCE. Participants were shown well-known

objects and materials (e.g. brick, or sand). The stimulus was removed, and then the experimenter

asked for immediate and direct color recall of what had been seen via selection from among 20

Munsell patches: “The observer viewed nearly the entire gamut of the Munsell 20-hue large book

samples displayed simultaneously on a large table, and he was asked to select that sample which

best represented the colour of a familiar absent object.” (Newhall et al. 1957, p. 52). Responses

were drawn to prototypical color exemplars —to individual examples that are better members of
Evidence for the Memory Color Effect 12

their category than others. This result was interpreted to reflect an association between canonical

colors and category prototypes. But no control examined whether such an effect occurs without

canonical and named stimuli. Perhaps people are generally biased to remember colors as more

prototypical. Recent and past evidence shows that they are (Bae et al., 2015). Thus, this study may

reflect a general color bias, as opposed to one applying to objects with canonical hues. Moreover,

because the methods relied on recall from memory, the results cannot be interpreted as

demonstrating an effect on appearance as opposed to memory. The study can therefore be

categorized as one that suffers from ambiguity with respect to appearance, allowing for an effect

on what people remembered, not what they saw. And finally, the study sought a continuous effect

in the form of stronger effects when an object was both canonical, and explicitly named, as opposed

to only canonical, comparing for example a condition in which participants were told to remember

the color of this object (and shown a brick) vs. told to remember the color of this brick. But as in

Bruner’s experiments, relative modulation was not observed.

Bolles et al. 1959. Like the study by Fisher and colleagues (1956) these authors questioned

the validity of past effects, focusing on the issue of response bias. They compared conditions in

which the correct color match was available as an object for selection —the ‘Satisfactory

Condition’— with a condition in which two options available to the observer did not include the

actual match —the ‘Unsatisfactory Condition.’ Like Newhall and colleagues, this study used a

direct recall paradigm. But in this study, no MCE was found in the Satisfactory condition, and it

was only found in the Unsatisfactory condition. In other words, subjects were only biased to a

canonical hue if the correct hue responses were unavailable among their choices. The experiment

is thus positive evidence for the possibility of accommodation, and it fails to replicate previous

evidence of the MCE since the failed Satisfactory condition in this study was equivalent to the
Evidence for the Memory Color Effect 13

condition with an effect in the study by Newhall and colleagues. Why would subjects show bias

in the condition with the right answer unavailable, but not when it was? We must assume that

participants did not misremember the colors that they saw, since they could answer accurately

when it was possible. That they showed biases when it was not possible must therefore indicate

some implicit or explicit awareness of the biases that would be relevant in an experiment like this.

This may be a case were the presence of two conditions produced accommodation in the condition

that appeared to invite it more directly. This is among the least cited of the studies from the 1950s.

Bartleson, 1960. This study and subsequent follow ups by the same group were very

similar to the study already described by Newhall and colleagues (1957; see also Bartelson 1961;

Bartleson and Bray 1961a). Here, a set of canonically colored stimuli were shown, verbally named,

and then participants were asked to recall the hue that the stimulus was shown in, choosing from

a set of 931 Munsell patches, in this case. Following previous methods, the stimulus was shown in

its canonical color, but an atypical, desaturated version thereof. Participants recalled the colors as

more saturated than actually shown. Here again, drawing conclusions about a bona fide MCE is

limited by the presence of the problem of general color bias, as well as ambiguity with respect

to appearance caused by the inherent memory component of the experiment.

Interim Summary. We have now discussed eight different studies spanning the dates 1923

to 1960. The studies from this period were characterized by direct report methods, experimenter

intermediation, experimental transparency, and often a focus on a single color category effect.

The studies, as a result, a prone to interpretational challenges caused by the possibility of subject

accommodation, remaining ambiguity with respect to appearance, and general color biases. In

several places, as well, data reporting standards fall short, failed predictions and replications

complicate the literature as a whole, and continuously or even conditionally modulated effects fail
Evidence for the Memory Color Effect 14

to obtain. We now turn to a second period of research, starting in 1965, when several innovations

begin to appear, particularly with respect to the replacement of direct report methods.

Delk and Fillenbaum 1965. This is perhaps the most well-known memory colour study

before the year 2000, cited over 120 times. In this study, subjects were individually presented with

either red diagnostic stimuli (e.g. lips, heart, apple), neutral colour diagnostic stimuli (e.g.

mushrooms, bells) or control stimuli (e.g. ellipse). Irrespective of whether a given stimulus

possesses a memory colour or not, each one was presented in an orange-red hue (Munsell chip

R/5/12) while contrasted against a reddish scaled surround. Subjects were able to adjust the

surround along a red-yellow-orange axis. The task was to adjust the surround until it exactly

matched the color of the object in foreground. A major accomplishment of the study was to

disambiguate effects of memory from appearance through the avoidance of direct report methods.

Participants were instructed to adjust the surround to match exactly what they were seeing in the

foreground rather than to report the foreground color through a match with a nonadjacent surface.

Because the foreground and surround were contiguous, successful adjustment should lead to the

foreground object vanishing, more or less. The logic was that participants would adjust the

background to match what they saw in the foreground, until no real boundary was apparent, even

if what they saw was not exactly the color presented. Would they adjust the surround to make it

redder for the canonically red objects?

Participants did adjust the background to a redder point for the canonical stimuli. The

major problem with the experiment, however, was that these adjustments were implemented by

the experimenter based on oral from the subjects, i.e. they were experimenter mediated. At least

one recent report failed to replicate these results, while using unmediated me (Gross et al., 2014).
Evidence for the Memory Color Effect 15

Leibovich and Paolera 1970. This study asked a slightly different question from the

others: Might the MCE manifest by producing a stronger after-image for canonically compared to

non-canonically colored objects. An after image is the appearance of an opponent color on a white

(or neutral) background after an observer fixates a color for a long enough duration. The basic

phenomenon can be experienced easily. Fixate the center of the red disc in Figure 2a, and after

about 10 seconds, quickly move your eyes to fixate the small black disc just to the right, in Figure

2b. You should see a bluish illusory disc against the empty white background.

Will a strawberry or a heart produce a more intense after image than a disc when each is

shown in the same color? If one perceives a strawberry as redder than a disc, all physical variables

held equal, then maybe a redder appearing strawberry should produce a more intense after image.

Leibovich and Paolera experimented with red and yellow canonically colored objects. But they

failed to obtain significant effects. Arguably, a mechanism for an MCE need not entail effects on

afterimages, which are usually thought of as low-level effects that depend only on physical

stimulus properties. But the study nevertheless adds some balance to the appearance of largely

positive evidence in the literature. Like other failed, but subtle predictions, it demonstrates that to

the extent that an MCE may exist, it is not obvious how it fits into a current and rather detailed

understanding of color perception.

Evidence for the Memory Color Effect 16

Figure 2A-D. Colour contrast demonstration. A. A circle disk with a fixation point directly

at its center; B. and D. with a fixation point on a neutral surround; C. A strawberry with a fixation

point directly at its center. By fixation on the center of A. or C. for at least 30 seconds and

immediately following at B. or D., one is said to experience an afterimage sensation.

White and Montgomery 1976. This study replicated the one just described, by Leibovich

and Paolera (1970), out of concern that the college students tested in that study would not yet have

“adult perception.” Participants in the age range of 19-23 were therefore tested, and in this study,

the American flag was used as the high colour diagnostic stimulus, while a random stripe image

with no identity connection to another flag was used as the control stimulus. Results showed that,

on average, subjects overestimated the hue of the afterimage for the American flag when compared

to the control stimulus. These results not only contradict those obtained by Leibovich and Paolera

(1970), but as White and Montgomery (1976) argue, suggest that the memory colour of diagnostic

stimuli is able to indirectly modulate perception in a mature sample population.

These are the only two after image studies that we are aware of, and so only further

experiments could adjudicate the disagreement directly. But we argue for skepticism in the case

of White and Montgomery for two reasons. First, it is not theoretically obvious why college age

students should not show the effect, if older adults do. Indeed, if that is the case, then there is

clearly something worth investigating to understand why. No obvious mechanisms jump out. We

see this as a cautionary sign, an overly specific effect without a mechanism.

Further, afterimages really are typically understood as relatively low-level phenomena, in

general, and in particular in the case of color (Stockman & Brainard, 2010), where cone activity

in the retina is the is the starting point for the effect. Even if we endorsed the presence of an MCE
Evidence for the Memory Color Effect 17

in general, our prior with respect to an effect on afterimages would be low. Is this then possibly an

example of the El Greco effect? The El Greco effect is a term recently employed by Firestone

(2013) to classify effects that appear in places where the experimenter should not expect them to,

effects that likely reflect some kind of accommodation on the part of participants. Recall that in

this study the stimuli were only an American flag, and a non-flag arrangement of stripes in the

same colors as the American flag shown.

Siple and Springer 1983. This study did not seek to show an effect of color memory on

color appearance. But we describe it here nonetheless because it appears to occasionally be cited

as evidence of such, and because its actual findings are instructive with respect to studies on

memory and appearance. Siple and Springer sought to answer a straightforward question: To the

extent that that people remember the colors of foods that they have seen in the past, do they

remember them accurately. Does one’s memory for the yellow of bananas match the actual yellow

of most bananas? To answer the question, they photographed real fruit and confirmed with

technical rigor that the colors of the fruit in their pictures matched the actual colors of the fruit in

a produce store. Then they removed the color from the images, and with a colorimeter, they asked

participants to adjust the colors of the images to match the ones that they remember the respective

foods possessing on a typical day. There were several conditions in the experiment that go beyond

the scope of this review; suffice to say that the result was that participant hue and brightness (value)

estimates were strikingly accurate, compared with the actual measured colors of the foods in the

images and in reality. But estimates of saturation (chroma) were biased such that foods were

remembered as more saturated than they actually tend to be. Two key points emerge for current

purposes: first, the study did not include any results that were meant to be interpreted as or can be

interpreted as evidence of an effect of memory on appearance. But the study is at times cited as
Evidence for the Memory Color Effect 18

such. Second, the results further suggest that general biases in color estimation may be pervasive

and systematic (see also, Forder & Lupyan 2017).

Second Interim Summary. The second wave of research on the MCE can be characterized

by an attempt to venture away from direct reports of perceived color. The study by Delk and

Fillenbaum did so by employing a boundary matching technique that presages contemporary

research which emphasizes achromatic identification (described below). But it suffers from among

the most concerning methodological problems, experimenter intermediation. Subsequent studies

avoided direct report by looking for an effect of perceived color on perceived after image. This

clever approach leverages broader knowledge of color perception phenomena, but it produced

conflicting results that appear not to have been the subject of further investigation.

Studies 2006- 2016

Studies by Hansen, Olkkonen, Gegenfurtner, Witzel, and colleagues. Starting in the

early 2000s, several color vision experts began to co-publish a series of papers with positive

evidence for the MCE. Because of similarities in methods across these studies, conducted largely

by the same group of individuals, we will not review these papers exhaustively, and instead, we

describe the major methodological innovations in these papers and their findings, followed by

potential concerns. The relevant papers are as follows: Hansen et al. 2006; Olkkonen et al. 2008;

Mitterer et al. 2009; Witzel et al. 2011; Kimura et al. 2013; Vurro et al. 2013; Witzel 2016.

The first major innovation in these contemporary studies is the use of rigorously controlled

stimuli and computerized responses produced by participants (i.e. without intermediation) in

established color spaces (usually DKL). The second major innovation involves the specific task in

the critical task given to participants: to adjust a shown stimulus until it appears achromatic, that
Evidence for the Memory Color Effect 19

is a neutral gray. For example, a participant might be shown a banana, in a random color, and she

would be asked to adjust the color of the banana (through keypresses that walk through the color

space) until it appears to her as a neutral gray (Figure 3)

Figure 3. Hue scaling method in DKL color space. The subject is presented with a banana

in a random hue. They are instructed scale the surface banana until it appears gray indicated by the

arrow.

The logic is that for canonically colored objects, the MCE should cause truly achromatic

stimuli to appear slightly chromatic. Adjustment to a point that appears achromatic should

therefore require adding slightly opponent color to gray. So, a banana’s achromatic point should
Evidence for the Memory Color Effect 20

be slightly blue, whereas a control discs point should be more genuinely achromatic. In other

experiments, the same logic has been applied in a comparative paradigm: participants are asked

which of two stimuli appears more neutral, with one banana, for example, being genuinely neutral,

and the other choice being slightly blue. The prediction is that the slightly blue-gray banana should

be selected as opposed to the truly gray one. Both kinds of results have obtained in the cited papers.

Bananas, and other objects appear to be achromatic in appearance only when they possess small

amounts of their opponent-from-canonical colors.

These are compelling results, heavily cited, appropriately held up as the best evidence of

the MCE, and there are roughly half a dozen or so relevant replications in the cohort of papers.

But because these papers all share much of the same methodology, there is the possibility that a

confound could apply broadly. We describe here some potential concerns that have risen to our

attention, with suggestions for how the basic methods could be made to be more divers, thereby

supplying converging and more compelling evidence still.

Our first cause for concern is the fact that in the best of these studies (Hansen et al., 2006;

Olkkonen et al., 2008; and Witzel et al., 2011) the achromatic locus for canonical stimuli is largely

in the blue region of color space (possibly suffering from the single-color category effect).

Another way to put this: the effects seem to be most easily obtainable with yellow canonical

objects, yellowish-green ones, and orange ones, almost never with red or blue ones. Are bananas

more canonically yellow than blueberries are blue? Perhaps. But the results would certainly be

more compelling if effects could be obtained with objects whose opponent achromatic point were

yellow or even orange or green. This is especially true because it is known that there is a general

bias to see blueish gray as more firmly neutral than other tinted grays, reddish gray for example.

A related cause for concern is that the degree of variability in the hues added to achromatic settings
Evidence for the Memory Color Effect 21

appears to vastly underestimate the variability in the actual colors of canonical objects. This further

suggests a general achromatic bias as the source for the main opponent effect, as opposed to one

that is driven by the canonical hues of the relevant objects. It has been suggested in some of the

relevant papers that the lack of an effect with objects that should produce red-opponent achromatic

points can be explained because the effect will only arise for opponent points that are on the axis

of daylight variability. But it is not clear why. The lack of an effect should not be written off with

emphasis only on the cases that do produce effects. Instead, a more thorough cataloging of effects

as a function of canonical color could empirically resolve the issue.

Could these experiments suffer from the persistent problem of accommodation, that

participants are to some extent aware of the purpose of the experiment and generating results that

do not reflect perception. Accommodation is perhaps to specific a term to apply to experiments

with this degree of rigor. The authors have fairly noted that it seems farfetched to think that naïve

participants realize that the name of game is to produce an opponent but still achromatic hue. Note

that in the experiments that make use of continuous adjustment, the objects start in random colors,

not their canonical ones.

We agree that it is unlikely that participants intentionally turn a red banana into a slightly

blue gray one on purpose, because they are aware of the experimenter’s goals. But participants

may not need to guess correctly to unintentionally or intentionally produce effects that do not

reflect the appearance of the objects at an achromatic point. For one, just as participants likely to

do not understand the purpose of the experiment, it is not obvious that the average subject will

understand the methods. If the instructions direct the participant to make the object look gray, well,

there are many different individual hues a lay person would call gray. It could even be the case

that participants understand the instructions to mean that each object should be made to look gray,
Evidence for the Memory Color Effect 22

but not that they should all be the same exact gray (which is the null prediction of experimental

logic). To a lay person, the incorrect expectation about the experiment may be that the goal is to

differentiate among the different objects used. Similarly, if the instructions are to make the

stimulus appear neutral or achromatic, it is not obvious that lay subjects would understand or

interpret these terms in their technical senses. It is even conceivable that a participant could

interpret the instructions to mean “adjust each stimulus there is no trace of what you think of as its

canonical color.” In such a case, neutral for a banana would mean “make sure it has not a trace of

yellow.”

What could be done to leverage the strengths of these studies and bolster their conclusions.

As we already noted, exploration of a wider canonical color gamut is a clear path. Certain features

of the data could be further explored as well, to determine if there are any signs of the kinds of

problems we have identified. For example, in the experiments that use a continuous adjustment

technique, the paths taken by participants to arrive at the neutral points could be informative. When

the object is a banana, do participants pass through yellow on the way to gray more often than

when it is a strawberry? The random starting points mean that they need not. But if they do, it

could reflect a strategy of adjusting out the canonical color, perhaps owing to a misinterpretation

of the instructions. More debriefing around the issue of how participants understand the

instructions could be useful as well. And above all, additional experiments with different methods

could do the trick. One approach would be to follow the suggestions made by Firestone and Scholl

(2015) to look for an effect where it should not appear in order to determine if participants are

(knowingly or not) creating the effects and not reflecting their perception. In this case, one could

ask whether the names of canonically colored objects produce the effects. Does the word banana

look more yellow than the word strawberry? It probably should not. Alternatively, something like
Evidence for the Memory Color Effect 23

the approach taken by Delk and Fillenbaum (1965) could potentially shore up these results (absent

experimenter intermediation, of course). Suppose the background on which the object appears is

the independently measured achromatic locus for a given subject. Now the instructions could avoid

grayness, neutrality and the like. Participants could just be instructed to adjust the color of the

object shown until they feel like they can’t see its outline at all against the background.

Conclusion. Do objects with canonical colors appear more canonically colored than other

objects, all else held equal. We would argue that the jury is still out. The recent evidence makes a

strong case for the existence of the effect. But as we just explained, we think it bears further

confirmation, particularly when considering the more recent study by Adeyefa-Olasupo (2018)

was unable to replicate the effects of Hansen et al. (2006) and found in the critical task, that

observers typically adjusted color diagnostic objects in the opponent direction of their achromatic

points. Moreover, if the evidence gets stronger, it will be time to uncover the unidentified

mechanisms by which such an effect could arise, given what we already know about color

perception. Finally, while this set of papers provide compelling evidence for the purported effect,

the historical literature falls short of contemporary standards in a variety of ways, when thought it

is frequently cited without caveats, and without accounting for the failed effects and replications

(of which there are likely more than have been published). The long history of papers available to

cite can make the effect seem like one with a long history of replication and convergence. But in

this case, we would argue that the effect’s history should be treated as no more than the runway to

the methodological rigor of contemporary experiments, of which more are needed for firm

conclusions to be drawn.
Evidence for the Memory Color Effect 24

References

Adams, G.K. (1923). An experimental study of memory color and related phenomena. The
American Journal of Psychology. 34 359–407.

Adeyefa-Olasupo, I E. (2018) Perception is shaped along the (L+M)–S axis (and possibly
confused with a memory color effect) bioArXiv Preprints.

Bae, G. Y., Olkkonen, M., Allred, S. R., & Flombaum, J. I. (2015). Why some colors appear more
memorable than others. A model incorporating categories and particulars in colour working
memory. Journal of Experimental Psychology: General. 144(4):744–63.

Bartleson, C.J. (1960). Memory colors of familiar objects. Journal of the Optical Society of
America. 50 73–77.

Bartleson, C.J. (1961). Color in memory relation to photographic reproduction. Phot Sci Eng,
Vol.5 No.6, p.327-331.

Bartleson, C.J., C. P. Bray. (1961). On preferred reproduction of flesh, blue-sky, and green-grass
colors. Phot Sci Eng, Vol.6 No.1, p.19-25.

Bolles, R. C., Hulicka, I. M., Hanly, B. (1959). Color judgment as a function of stimulus conditions
and memory colour. Canadian Journal of Psychology/Revue canadienne de psychologie 13
175–185.

Bruner, J. S., Postman L., Rodrigues, J. (1951). Expectation and the perception of color. The
American Journal of Psychology 64 216–227.

Delk, J. & Fillenbaum, S. (1965). Differences in perceived colour as a function of characteristic

colour. The American Journal of Psychology 78:290–93.

Duncker, K. (1939). The influence of past experience upon perceptual properties. The American
Journal of Psychology 52 255–265 Page 25 of 28 Attention, Perception, & Psychophysics

Durgin, F. H., Baird, J. A., Greenburg, M., Russell, R., Shaughnessy, K. & Waymouth, S. (2009)
Who is being deceived? The experimental demands of wearing a backpack. Psychonomic
Bulletin and Review 16:964–69.

Firestone, C., & Scholl, B. J. (2015). Cognition does not affect perception: Evaluating the evidence
for ‘top-down’ effects. Behavioural and Brain Sciences.1-71.

Firestone, C. (2013b) On the origin and status of the “El Greco fallacy.” Perception 42:672–74.

Fisher S C, Hull C, Holtz P, (1956). Past experience and perception: memory color. The American
Journal of Psychology 69 546–560.

Fodor, J. A. (1983) Modularity of mind: An essay on faculty psychology. MIT Press. Fodor, J. A.
Evidence for the Memory Color Effect 25

(1984) Observation reconsidered. Philosophy of Science 51:23–43.

Fodor, J. A. (1988) A reply to Churchland’s “Perceptual plasticity and theoretical neutrality.”

Philosophy of Science 55:188–98.

Forder, L., & Lupyan, G. (2017). Facilitation of color discrimination by verbal and visual cues.
PsyArXiv Preprints.

Granzier J, Gegenfurtner K, (2012). Effects of memory colour on colour constancy for unknown
colored objects. I- Perception; 1 (3) 190 – 215.

Gross S., Chaisilprungraung T., Kaplan E., Menendez J., Flombaum J. (2014). “Problems for the
purported cognitive penetration of perceptual colour experience and Macpherson’s
proposed mechanism,” in Thought and Perception, eds Machery E., Prinz J., editors.
(Lawrence, KA: New Prairie Press), 1–30.

Goldstone, R. L. & Barsalou, L. W. (1998) Reuniting perception and conception. Cognition

65:231–62.

Harper,R. S. (1953). The perceptual modification of colored figures. The American Journal of
Psychology, 66 86–89.

Hansen, T., Olkkonen, M., Walter, S. & Gegenfurtner, K. R. (2006) Memory modulates color
appearance. Nature Neuroscience 9(11):1367–68.

Hering E, (1920) [1878] Grundzüge der Lehre vom Lichtsinn (Berlin, Germany: Springer).

Hendley, C. D., & Hecht S. (1949). The colours of natural objects and terrains, and their relation
to visual colour deficiency. Journal of the Optical Society of America, 39(10), 870–872.

Heurley L. P., Milhau A., Chesnoy G., Ferrier L. P., Brouillet T., and Brouillet, D. (2012).
Influence of language on colour perception: a simulationnist explanation. Biolinguistics 6,
354–382.

Hanawalt N.G. & Post, B.E. (1942). Memory trace for colour. J. Exp. Psychol. 30, 216-227.

Heurley L. P., Milhau A., Chesnoy G., Ferrier L. P., Brouillet T., and Brouillet, D. (2012).
Influence of language on colour perception: a simulationnist explanation. Biolinguistics 6,
354–382.

Hurvich, L. M & Jameson, D. (1957). An opponent process theory of colour vision. Psychological
Review, 64, 384-404.

Klein, G. S., Schlesinger, H. J., & Meister, D. E. (1951). The eVect of personal values on
perception—an experimental critique. Psychological Review, 58, 96–112.

Kanizsa, G. (1985) Seeing and thinking. Acta Psychologica 59:23–33.

Evidence for the Memory Color Effect 26

Knill, D. C. & Richards, W. (eds.). (1996). Perception as Bayesian inference. Cambridge:

University Press.

Kimura, A., Wada, Y., Masuda, T., Goto, S., Tsuzuki, D., Hibino, H., Dan, I. (2013). Memory
color effect induced by familiarity of brand logos. PLoS One, 8, e68474.

Krauskopf J, Williams D R, Heeley D W. (1982). Cardinal directions of colour space. Vision

Research; 22:1123–1131.

Lupyan, G. (2012) Linguistically modulated perception and cognition: The label feedback
hypothesis. Frontiers in Psychology 3: Article 54.

Leibovich B, Paolera M B, (1970). Research on colour perception and after-image in relation to

significant images” Annales Médico-Psychologiques 128 735–742.

Marr, D. (1982). Vision: A computational investigation into the human representation and
processing of visual information. San Francisco, CA: W.H. Freeman.

Machery, E. (2015) Cognitive penetrability: A no-progress report. In: The cognitive penetrability
of perception: New philosophical perspectives, ed. J. Zeimbekis & A. Raftopoulos. Oxford
University Press. Page 27 of 28 Attention, Perception, & Psychophysics

Mitterer, H. & de Ruiter, J. P. (2008) Recalibrating color categories using world knowledge.
Psychological Science.19:629–634.

Mitterer, H., Horschig, J. M., Müsseler, J. & Majid, A. (2009) The influence of memory on
perception: It’s not what things look like, it’s what you call them. Journal of Experimental
Psychology: Learning, Memory, and Cognition 35:1557–62.

Newhall S M, Burnham R W, Clark J R, (1957). Comparison of successive with simultaneous

colour matching. Journal of the Optical Society of America 47 43–54.

Pylyshyn, Z. (1999) Is vision continuous with cognition? The case for cognitive impenetrability
of visual perception. Behavioral and Brain Sciences 22(3):341– 65.

Pylyshyn, Z. W. (2002) Mental imagery: In search of a theory. Behavioral and Brain Sciences
25:157–238.

Proffitt, D. R. (2013) An embodied approach to perception: By what units are Visual perceptions
scaled? Perspectives on Psychological Science 8:474–83.

Vurro, M Ling, Y., Hurlbert, A. (2013). Memory colour of natural familiar objects: Effects of
surface texture and 3-D shape. Journal of Vision 13(7):20, 1–20.

White C W, Montgomery D A. (1976). Memory colours in afterimages: a bicentennial

demonstration” Perception & Psychophysics 19 371–374.

Winkler A.D. Spillmann, L, Werner, J.S. & Webster, M.A. (2015). Asymmetries in blue-yellow
Evidence for the Memory Color Effect 27

colour perception and in the colour of 'the dress. Current Biology 25(13): pR547–R548

Witzel, C., Valkova, H., Hansen, T. & Gegenfurtner, K. R. (2011) Object knowledge modulates
colour appearance. i-Perception 2(1):13–49.

Witzel, C. (2016). An easy way to show memory colour effects. iperception 7, 1–11.