At a glance
The mere exposure effect explained through five landmark studies, from Zajonc's 1968 monograph and the subliminal octagon experiment to Bornstein's meta-analysis, the 2017 curve refit and salience theory.
In 1968 the psychologist Robert Zajonc published a monograph with a title that promised very little: “Attitudinal Effects of Mere Exposure.” Its argument was that people come to like a thing simply because they have encountered it before, with no reward attached and no argument made on its behalf. Usually there was no memory of the earlier encounter at all. Mere, in his usage, meant a condition that made a stimulus accessible to perception and nothing more.
Half a century of testing has not overturned the claim. It has narrowed it, priced it, and in the past few years begun to argue about what actually produces it. The five papers below are the ones researchers reach for when they need to show where the finding came from, how large it is, and where it stops. Read together, they describe an effect modest in size — meta-analytic summaries put the association between repetition and liking in the low-to-mid .2s on a correlation scale — and unusually wide in reach, from irregular polygons to the faces of classmates.
Why a modest correlation travels so far
An effect of that size does not change a single decision. It tilts the odds slightly, in the same direction, every time a person meets a brand, a candidate, a song or a colleague again.
The commercial arithmetic was spelled out in unusually blunt terms in an expert report filed in United States tobacco litigation, in which the psychologist Paul Slovic argued that cigarette advertising works partly by displaying a brand name or package that creates positive affect through the mere exposure effect. The report added that the resulting good feeling could depress the perceived risk of smoking.
Software has begun to encode the same premise. A 2023 paper presented at the ACM Conference on Recommender Systems described a model, Ex2Vec, that uses the exposure effect to characterise users and items and to predict future music consumption on a streaming service, on the assumption that interest rises across early repetitions and falls after a peak.
| Year | Paper | Journal | What it changed |
|---|---|---|---|
| 1968 | Zajonc, “Attitudinal Effects of Mere Exposure” | Journal of Personality and Social Psychology | Turned a scattered observation into an experimental paradigm |
| 1980 | Kunst-Wilson & Zajonc, “Affective Discrimination of Stimuli That Cannot Be Recognized” | Science | Showed preference could form where recognition failed |
| 1989 | Bornstein, “Exposure and Affect” | Psychological Bulletin | Measured two decades of experiments and mapped the moderators |
| 2017 | Montoya, Horton, Vevea, Citkowicz & Lauber, “A Re-Examination of the Mere Exposure Effect” | Psychological Bulletin | Fitted curves rather than averages; found limits by sense and duration |
| 2020 | Mrkva & Van Boven, “Salience Theory of Mere Exposure” | Journal of Personality and Social Psychology | Argued relative exposure, not raw frequency, does the work |
1968: four kinds of evidence in one place
The monograph is odd by present-day standards because it does not rest on a single experiment. Zajonc assembled four separate lines of evidence and argued that they converged. The first was linguistic: words carrying favourable meanings appear far more often in printed English than their opposites, a pattern he checked against the Thorndike–Lorge frequency counts of written usage.
The second was experimental, and it is the one that became the template. Participants saw Chinese characters, nonsense words or photographs of faces at controlled frequencies, between one and twenty-five times, then rated each on a seven-point scale anchored at good and bad. The characters worked because they were visually distinctive and semantically empty for the American undergraduates in the study: no prior meaning could explain why one ideograph was preferred to another. Ratings rose with frequency, and for faces the rise was close to linear. A third strand tied word frequency to attitudes about what words refer to; a fourth reported direct manipulations of exposure.
Zajonc also weighed rival accounts, including exploration theory and the possibility that repeated viewing produces semantic satiation rather than liking, and set them aside with the evidence then available. The contribution was less the discovery of a new phenomenon than the conversion of scattered observations into a paradigm with a standard procedure, a standard measure and a standard result. Nearly every study that follows uses a version of it. The American Psychological Association's record of the monograph summarises the four evidence streams in its abstract, which is how most students first meet them.
1980: liking what you cannot identify
If preference needs no reward, does it need awareness? William Raft Kunst-Wilson and Zajonc, then at Michigan, put the question in Science. The stimuli were twenty irregular octagons generated at random and therefore free of associations. Each participant saw five presentations of ten of them at one millisecond each, through a viewing tunnel with the illumination cut and a red filter over the projector, the exposures degraded until pilot work showed recognition would fall to chance.
Recognition did fall to chance: 48 per cent in paired comparisons run at a full second. Preference did not. Previously seen octagons were chosen 60 per cent of the time, a result the authors placed at p < .01. Sixteen of the twenty-four participants preferred the old shapes overall, while only five could pick them out better than chance. Seventeen of the twenty-four discriminated more accurately with their liking judgments than with their memory. The two-page paper is among the most cited in the exposure literature, and among the most quarrelled with.
Sceptics have a case. Two pages in 1980 cannot settle the boundary between perception and awareness, and the subliminal literature built on this design has had a difficult replication record; a Swinburne doctoral project built a purpose-made LED tachistoscope after finding that even one-millisecond presentations could be recognised well above chance. It does not follow, however, that the asymmetry Kunst-Wilson and Zajonc reported was an artefact. Bornstein's later review found exposure effects were larger, not smaller, in studies where recognition failed, which is the opposite of what a pure measurement error would produce.
Photo by Andre William on Unsplash
1989: counting the literature instead of quoting it
By the late 1980s the exposure literature had grown large, uneven and easy to cherry-pick. Robert Bornstein's meta-analysis covered the two decades of experiments that followed Zajonc's monograph and asked which design choices changed the size of the effect. Almost all of them did: stimulus type, stimulus complexity, whether presentations were blocked or mixed, exposure duration, whether the stimulus was recognised, the age of the participant, and the delay between exposure and rating.
Two findings did more work than the rest. The effect peaked early. Experiments using relatively few presentations produced larger effects, those running past a hundred presentations much weaker ones, and the relation between frequency and liking was positive without being linear. A ceiling in reported liking was reached after a small number of viewings. And the effect held across every class of stimulus he examined — polygons, photographs, paintings and drawings, ideographs, nonsense syllables, meaningful words, real people — with combined effect sizes clustering in the low .2s. He used the pattern to revise the two-factor model of the effect and floated an evolutionary reading: an organism that feels mildly positive about the familiar is, on average, safer than one that does not. The full record of the review remains the standard citation for the effect's magnitude.
2017: the curve, redrawn and qualified
R. Matthew Montoya and four colleagues re-analysed the field with a different technique, fitting curves rather than averaging effect sizes: 268 curve estimates drawn from 81 articles. The average shape was an inverted U, a positive slope and a negative quadratic term, rising across early repetitions and turning down later. That confirmed the older picture.
The qualifications were new. The curve appeared for visual stimuli but not auditory ones. It appeared for exposures shorter than ten seconds and, separately, for exposures longer than a minute, an odd pair of conditions unless the two ends engage different processes. And the authors concluded that the standard explanations, processing fluency and reduced apprehension, could not account for what the data showed. The abstract and citation record set the terms for most work done since.
The inverted U is not only a laboratory shape. In a study Richard Moreland and Scott Beach published in 1992, four women of similar appearance posed as students in a large lecture course, attending no sessions, five, ten or fifteen across a term, and never interacting with anyone. At the end of the semester 130 students rated slides of the four. Attractiveness scores climbed with attendance, from 3.62 to 4.38 on a seven-point scale. Recognition barely moved: students liked the women they had seen more often without reliably remembering that they had seen them at all.
2020: exposure, or standing out?
The most direct challenge to the standard account came from Kellen Mrkva and Leaf Van Boven, who proposed that repetition works by making a stimulus salient — the quality of standing out against competing stimuli — rather than by making it easier to process. Salience is inherently relative. A rare car is salient on an empty motorway only to someone who collects cars.
Their experiments first reproduced the familiar result: stimuli shown nine times were liked more than stimuli shown three times, once or not at all, and were also rated more salient, more extreme and more emotionally intense. Mediation models and a chain of experiments traced liking back to salience, with fluency and apprehension explaining less of it. Then came the decisive test, manipulating relative and absolute exposure separately. Stimuli presented nine times were liked more when everything else in the set had appeared fewer than nine times than when other stimuli had appeared more often. In that experiment absolute exposure had no significant effect at all. The published study runs to twenty-eight pages of experiments making the case.
If that holds, the practical rule changes. There is no fixed number of repetitions that breeds liking, only a comparison between this stimulus and its neighbours.
Voices are where the field has recently run into trouble
The boundary conditions are now the interesting part of the field, and several are recent. Kalliopi Mavromati, Hyuner Erdinch and Helena Paterson tested Scottish voices with 238 participants listening to spoken sentences and 88 listening to single words. Voices heard eight times were liked more than unfamiliar ones. Voices heard twice or five times were liked less than unfamiliar voices, and were recognised with less confidence, forming what the authors call an uncanny valley of familiarly unfamiliar speakers. Listeners' prior experience with the accent moderated the pattern.
The effect also survives a change of medium. A 2024 study by Daniel Hepperle and Matthias Wölfel replicated Mrkva and Van Boven's design remotely on computer screens and in immersive virtual reality. Liking rose with exposure in both settings, results were statistically similar between them, and both aligned with the 2017 meta-analysis.
And the effect is not confined to the stimulus. In 2000 Jennifer Monahan, Sheila Murphy and Zajonc reported that participants given repeated five-millisecond exposures rated their own mood more positively than participants given single exposures, and that the resulting affect spilled over. Previously exposed stimuli were rated most positively, novel but similar stimuli next, and stimuli from an unrelated category were lifted too. That diffusion is what makes the effect commercially useful and ethically awkward.
Photo by The 77 Human Needs System on Unsplash
Who decides what a person meets twice
The precedent these papers set is procedural. Zajonc's generation established that a mental process could be studied through the residue it leaves on ratings of meaningless shapes. Bornstein's established that the residue could be counted across a literature rather than asserted by whichever study came to hand. Montoya's established that the literature had to be fitted as curves rather than averaged. Mrkva and Van Boven's established that the comparison set, not the count, may be the active ingredient. Each step made the effect smaller in its claims and larger in its consequences.
What none of them settles is consent. If liking is partly a function of relative frequency, then whoever controls frequency controls a small lever on preference, and the modern controllers are recommendation systems, advertising buyers and the interfaces that decide which faces, songs and arguments a person meets twice. That a public which never sees the control condition can still be said to have chosen its tastes is an assumption, not a finding. It is the assumption the next paper in this line will have to test.
- social-psychology
- social-psychology
- mere-exposure-effect
- mere-exposure-effect
- robert-zajonc

