Eye Contact: What It Predicts, What It Creates, and How Long Is Too Long

Two different claims, and only one of them is about reading anybody

Advice about eye contact runs two claims together without ever separating them. The first is that holding someone's gaze reveals how they feel about you. The second is that holding someone's gaze produces how they feel about you. They point in opposite directions — one treats the look as a readout, the other as an input — and the evidence for them is not the same evidence.

The readout claim has been tested at scale. Montoya, Kershaw and Prosser pooled 309 effect sizes from 54 studies and 5,422 participants and found eleven behaviours significantly associated with attraction. Gaze duration and gaze frequency are two of them. The overall association between enacted behaviour and reported attraction was z = .20, which is real and small, and the behaviours they call direct — physical proximity, the amount of talking, the duration of talking — related more strongly than the indirect ones that this genre is almost entirely about.

So looking is on the short list. The part that gets dropped is the author's own reading of what that list is. Montoya describes the eleven as trust-and-rapport behaviours: the things people do to get someone to trust them, and the same set that shows up between doctors and patients, parents and children, salespeople and customers. They are evidence of liking. Nothing in the analysis separates one kind of liking from another, which is the wall that telling flirting from ordinary friendliness runs into and the reason the wider body-language literature supports so much less than it is sold as.

The input claim is narrower and stranger. There is an experiment behind it, it is from 1989, it has two studies in it, and what gets quoted from it is roughly a third of what it found. The next two sections are about the rest.

The 1989 experiment the whole genre rests on

Joan Kellerman, James Lewis and James Laird published Looking and Loving in the Journal of Research in Personality in 1989. It is the source for every claim you have read that staring into a stranger's eyes creates romantic feeling, and it deserves to be described properly rather than summarised into a slogan.

Study I used 96 undergraduate volunteers, 48 men and 48 women, recruited for a study of the acquaintance process and paired with an opposite-sex stranger. Each member of a pair was briefed separately, so neither knew what the other had been told. Then they sat for two minutes in one of five conditions: both looking at the other's eyes, both at the other's hands, one at eyes and the other at hands either way round, or — the condition that makes the study worth reading — one of them counting the other's eye blinks.

Afterwards they rated their feelings on twelve bipolar scales. The affection scores came out at 62.4 for the pairs who had both been gazing and between 52.8 and 55.8 for everything else. The only significant effect in the analysis was the gaze condition, F(4, 80) = 2.50, p < .05, and it was simple: the mutual-gaze cell differed from all four of the others, and none of the others differed from each other. Being looked at was not enough. Looking was not enough. The eye-blink condition is the elegant part — objectively the same looking into someone's eyes, understood by the person doing it as counting — and it did not produce the effect either, differing from the mutual-gaze cell at t = 2.11, p < .05.

Study II added 72 more participants, 36 and 36, under a cover story about ESP, with the love and liking measures taken before and after. In the mutual-gaze condition, reported passionate love rose by 25.6 points, F(1, 56) = 44.52, p < .01, with dispositional love up 11.9 and liking up 10.2. The hand condition moved nothing in anyone.

That is the finding. It is one marginal analysis of variance on 96 people and one large within-group change in 72 more, published thirty-seven years ago, with no replication in the modern literature on gaze.

The half of that study nobody quotes

Two things sit in the same tables as the famous result, and neither has ever made it into an article about eye contact.

The first is that the Study II effect was a subgroup effect, and the paper says so in its own abstract. After the gaze session, participants did a facial-expression task that sorted them into people who take their emotional cue from their own behaviour and people who take it from the situation around them. The 25.6-point rise in passionate love belongs to the first group. The second group moved 6.8 points, which was not significant, and on the liking measure they went down by 11.1, F(1, 56) = 5.56, p < .05 — which the authors attribute to regression from higher starting scores rather than to the gaze. So the honest sentence is not that two minutes of eye contact makes strangers fall in love. It is that it raised a questionnaire score in roughly half of a sample of 72 undergraduates who had been told they were testing extrasensory perception.

The second is the room. Half the pairs sat in a dimly lit room with Keith Jarrett's Köln Concert playing softly; the other half sat under normal lighting in silence. The interaction between that and the before-and-after scores was significant, F(1, 56) = 11.27, p < .01. In the romantic room, passionate love rose 16.7 points in one cue group and 14.3 in the other, significantly in both. In the plain room it rose 6.0 and fell 1.3, significantly in neither.

Read those two results together. The lighting and the record moved the measure in everybody. The gazing moved it in half the sample. The study that is cited to prove eye contact creates love contains, in the same analysis, better evidence that low light and a piano record do — and the obvious reading of that pair is not that staring works but that a stranger's self-report about passionate love after two minutes in a laboratory is sensitive to the furniture. The same caution applies to anything you conclude about an evening that happened to be dimly lit, which is a cousin of the problem described in what actually builds attraction over time.

Where the four minutes came from, and it was not Aron

The version in circulation has a different number in it. Four minutes of unbroken eye contact, usually attached to Arthur Aron's questions and usually described as the last step of an experiment that made two strangers fall in love.

Aron's published procedure does not contain it. What Aron, Melinat, Aron, Vallone and Bator ran in 1997 was 45 minutes of conversation: three sets of questions that escalate in how personal they are, worked through by pairs of strangers, with matched small talk as the control. There is no staring task in the method. The outcome measured was closeness, not love, and the paper is more interesting than its reputation — we go through what it did and did not establish in what questions can and cannot reveal about compatibility, including the second study nobody cites.

Kellerman's gaze was two minutes, not four, and it was run without any questions at all. Nobody has published an experiment that combines the two, which means the four-minute figure is not an exaggeration of a result. It is a number that appeared in the retelling and then hardened, which is the same thing that happened to the three-day rule — invented at a table in a 1996 film and now quoted as though it had a source, as the piece on that rule sets out.

If you want to do the exercise, do it. It is harmless and people enjoy it. Just notice what it is: a shared awkward task with a stranger, which is a thing that reliably produces closeness for reasons that have nothing in particular to do with eyes.

Three and a quarter seconds, and what does not change it

There is one good modern number in this area, and it comes from the largest study anybody has run on the question of how much eye contact feels right.

Nicola Binetti, Charlotte Harrison, Antoine Coutrot, Alan Johnston and Isabelle Mareschal recruited 498 visitors to the London Science Museum — 224 men, 274 women, 56 nationalities, aged 11 to 79 with a mean of 29.9. Each watched video clips of eight actors looking straight at them for somewhere between 0.1 and 10.3 seconds, and after each clip said whether the look had felt too long or too short. Fitting a curve to those judgements gives the duration that felt right, and 428 participants produced usable fits. The average was 3,295 milliseconds, with a standard deviation of 706.

A bit over three seconds, then, for one unbroken look. What is more useful is the list of things that did not change it. The pairing of genders made no difference: male watching male, male watching female, female watching male and female watching female came out statistically indistinguishable, F(3, 424) = 1.45, p = .23. None of the five personality traits predicted it. And how attractive the participant rated the actor did not predict it either, which is worth sitting with, because the whole genre assumes that interest is what lengthens a look.

One face rating did predict it, and it was threat: the more threatening participants rated an actor, the shorter the gaze they wanted, r = −.13, p = .005. Small, but pointing the right way. The thing that did predict preferred duration strongly was outside anyone's control — people who preferred longer gaze showed faster increases in pupil size while watching, which the authors read as a difference in autonomic arousal.

The caveats are real. These were clips on a monitor at 57 centimetres with the head in a chinrest, in a museum, and the judgement was binary. It measures one look, not the share of a conversation. But as an order of magnitude it beats every number this genre circulates, and the spread is the point: there is no correct duration for a specific person, and the figure most likely to be wrong is a confident one. More of this kind of thing under Body Language.

A photograph does not look back, and most of the research is photographs

Almost everything written about the power of eye contact rests on studies in which participants looked at a picture. There is a good reason to think those studies are measuring something smaller than the thing they are used to explain.

Jari Hietanen, Jukka Leppänen, Mikko Peltola, Kati Linna-aho and Heidi Ruuhiala sat participants in front of another person and varied that person's gaze. Direct gaze produced relatively greater left-sided frontal EEG activity — the pattern associated with approach motivation — while averted gaze produced the right-sided pattern associated with avoidance. Skin conductance was higher to faces than to objects, and higher to direct than to averted gaze. Then they ran the same faces as photographs on a monitor, and the effects disappeared. Later work from the same laboratory found the same split in the brain's face-sensitive responses.

The clean test came in 2015. Aki Myllyneva and Hietanen seated a participant opposite a model behind an electronic shutter. In one condition the participant believed the model could see them through it; in the other they believed the pane was blocked from their side. The autonomic and brain responses to eye contact appeared only when the participant believed they could be seen. In a second experiment the model wore three pairs of sunglasses: empty frames, normal dark lenses — eyes hidden, but the model can still see — and identical dark lenses blacked out from the inside. The effect survived the eyes being hidden. It did not survive the model being unable to look back.

That is as close to a mechanism as this literature gets, and it is not the one the genre assumes. The active ingredient is not the visual appearance of two eyes pointed at you. It is the belief that another person's attention is on you. Atsushi Senju and Mark Johnson's review calls the general phenomenon the eye contact effect — perceived eye contact changing whatever processing is going on at the time — and it is worth noticing that the word doing the work in that definition is perceived.

It also carries a practical consequence that nobody draws. A face looking out of a photograph, a profile picture, a camera lens on a video call — none of these is the thing that was measured, because none of them can see you. Whatever a dating-app photo with direct gaze is doing, it is not doing this.

What people say about it and what their bodies do disagree

Hietanen reviewed the whole affective literature on eye contact in 2018 and found a split that is unusual enough to be worth knowing about before you interpret your own experience of a charged moment.

The explicit side — asking people how pleasant it felt — is inconsistent. Still photographs with direct gaze get rated as more pleasant than averted ones. Animated characters shifting their gaze give no effect at all. And when the stimulus is a live person sitting in front of you, direct gaze gets rated as slightly less pleasant than averted gaze or closed eyes, in study after study from that laboratory.

The implicit side is not inconsistent. Skin conductance, frontal EEG asymmetry, the startle reflex and the activity of the smiling muscle all line up: direct gaze raises arousal, and the reaction it produces is positively valenced and approach-related. The most quotable version is Knut Kampe, Chris Frith, Ray Dolan and Uta Frith's one-page report in Nature: activity in the ventral striatum, a region tied to reward prediction, rose with how attractive a face was rated when that face met the viewer's eye, and fell when the same face looked away. Reward systems appear to be involved in the opening moves of a social interaction — which is a long way from the claim that you can read somebody by watching theirs.

The pattern the review draws out of that is the useful part. The more a stimulus resembles actually being looked at by a person, the less positive people's conscious report of it becomes. Being attended to is arousing and, at a level people do not narrate, pleasant. At the level they do narrate, it is self-conscious.

So the sentence this page exists to undermine — it felt so intense, there must have been something there — is making two mistakes at once. Intensity is what the situation produces, not what the other person feels; and your own report of how it felt is the least reliable measurement in the experiment. That gap between an aroused state and the story told about it is the same machinery that makes love and anxious attachment so hard to tell apart from the inside, and our attachment style test gives a rough read on which way you tend to lean in about two minutes.

The same look reads as dominance, and sometimes as threat

When Chris Kleinke organised two decades of gaze research for Psychological Bulletin in 1986, he sorted it into five functions. Gaze provides information, regulates interaction, expresses intimacy, facilitates tasks — and exercises social control. That last one has been in the literature from the beginning. It is only in the advice that looking is purely affiliative.

The numbers back it. In Shota Uono and Hietanen's cross-cultural study, faces with direct gaze were rated as more dominant than the same faces with averted gaze, F(1, 58) = 42.21, p < .001, and that held across both groups of participants. In Binetti's museum sample, the one face rating that predicted how much gaze people wanted was threat. In Hironori Akechi's study, described in the next section, Japanese participants rated a neutral face making eye contact as angrier and less approachable than Finnish participants did.

Which is why this page will not tell you to hold someone's gaze a beat longer than feels natural, and will say the other thing plainly. A look that is held past the point where the other person has tried to break it is not a romantic technique with an upside. It is one of the ordinary ways intimidation is done, and it is the same behaviour whether it happens across a bar or on a train. If looking is arriving with being followed, being cornered, being pressed after a no, or being watched in a way that makes leaving feel difficult, none of that is on the spectrum this article describes. The helplines at the top of this page and in the sources are free, confidential and take about a minute.

The milder version matters too, because it is far more common. Monitoring somebody — who they looked at, how long, with what expression — is not evidence-gathering; it is the behaviour described in what surveillance of a partner actually does. And the general rule from the reading side holds here as firmly as it does in what the evidence supports about reading someone's interest: ambiguity is not an invitation.

How long is comfortable depends on where you learned it

Hironori Akechi, Atsushi Senju, Helen Uibo, Yukiko Kikuchi, Toshikazu Hasegawa and Hietanen ran the comparison properly. Twenty Finnish and twenty Japanese university students, ten women in each group, matched on age and on two measures of social anxiety, each facing a live model from their own cultural background through a liquid-crystal shutter while heart rate and looking time were recorded.

The physiology came out the same in both groups. Direct gaze produced a stronger heart-rate deceleration — the signature of attention orienting — shorter looking times when participants controlled the shutter themselves, and higher self-rated arousal, in Finnish and Japanese participants alike.

What differed was the evaluation. Japanese participants rated the direct-gaze face as angrier, less approachable and less pleasant than Finnish participants did, and rated it as angrier and sadder than the same face with averted gaze. The authors' conclusion is the one that matters here: the well-documented differences in how much eye contact people from different cultures make look like differences in display rules and what the behaviour is taken to mean, not differences in how the nervous system responds to being looked at.

A companion study pushed it further back. Uono and Hietanen showed 30 Finnish and 30 Japanese participants faces at eleven gaze angles, 176 trials each, and asked a question with no interpretation in it at all: is this face looking at me? Finnish participants were more willing to count a slightly averted gaze as aimed at them when the face was Japanese than when it was Finnish, F(1, 29) = 12.30, p = .002, while the Japanese participants showed no such difference. Even the perceptual judgement carries a bias, and the bias shifts with whose face it is.

Two student samples from two countries is a signal and not a map of the world; both papers say so. It is enough, though, to retire the instruction to maintain eye contact as if it were a constant, and enough to make the reading error obvious. A person raised to treat a long look as rude is not telling you anything about how they feel when they break it.

And on neurotype, which this genre never mentions

Dominic Trevisan, Nicole Roberts, Chantal Lin and Elina Birmingham did something the experimental literature had not: they went and read what autistic people say about eye contact. Their 2017 analysis covers 698 separate passages from 364 individuals, taken from ten YouTube videos and from the WrongPlanet forum. The diagnoses were self-declared and could not be verified, which the authors state.

The categories are not subtle. Twenty-two passages describe eye contact as physically painful. Twenty-four describe a fight-or-flight response — one of them, being spotted as prey. Fifty-four describe a sense of violation or exposure. And the single largest category, with 81 passages, is not about discomfort at all: it is being unable to process what somebody is saying while looking at their eyes. One participant gives a number that sits oddly close to the museum study's average, from the other side of it: the longest they can hold a gaze is two to six seconds before it becomes tiring. The most common coping strategy, 67 passages, was fixating somewhere else on the face.

There is physiological work pointing the same way, and it is small. Miia Kaartinen and colleagues measured skin conductance in 15 children with an autism diagnosis and 16 without while each looked at a live face. Among the autistic children, the more arousal direct gaze added over averted gaze and closed eyes, the greater their rated difficulties with language, communication and gesture. There was no such association in the comparison group. Thirty-one children in total: treat it as a lead rather than a result.

And it is not only autism. Andero Uusberg, Jüri Allik and Hietanen found that the frontal EEG asymmetry to a live person's direct gaze reverses with neuroticism — approach-sided in people scoring low, avoidance-sided in people scoring high.

Two things follow. The advice to hold eye contact longer has a price that the advice never quotes, and for some people the price is the conversation itself. And from the other chair: someone looking away may be carrying a load you do not have. The absence of eye contact is close to the least informative observation available, which is also where the evidence on reading a quiet person ends up.

What this actually leaves you with

Start with the gap, because it is larger than anything on this page. No study in the gaze literature, and none in the speed-dating work that codes behaviour, has measured how much eye contact somebody made on a date against whether the other person wanted a second one. The question the genre answers most confidently is the one nobody has asked.

What has been measured is how badly interest gets read, and eye contact is squarely inside that. Observers judging speed daters in Place, Todd, Penke and Asendorpf's study read men's interest well above chance and women's barely above it, with the errors running systematically towards seeing interest that was not there. In Hall, Xing and Brooks's recordings, people sitting three feet apart for ten minutes identified their partner's flirting 28% of the time it happened. Mason, Tatkow and Macrae did find that a face whose gaze shifted towards the viewer was rated more likeable and more attractive than one whose gaze shifted away — but only when the judgement was one the participant had a stake in, which is a narrower result than the title suggests. And Riechelmann and colleagues found that even the processing advantage for direct gaze is conditional: in a masked discrimination task, averted gaze was identified faster and more accurately.

So here is the whole of it. Around three seconds is a reasonable order of magnitude for a single unbroken look, with a wide spread and no right answer for a particular person. Comfortable looking arrives in bursts with natural breaks in them. Most of what makes a held gaze feel like something is that you are being attended to, which is also true of a good interviewer, a hairdresser and a doctor.

The one practical suggestion here is the inverse of the usual one: do not run it as a technique. Deliberately manufacturing a long look raises arousal, arousal is not information, and you will have produced the signal you were planning to read. What is left is unglamorous and works — look at someone while they are talking, let your eyes go where they want between sentences, and settle the question of whether they are interested by asking it, which is the conclusion the whole Signs shelf keeps arriving at and the thing research on how people open up supports better than any cue on any list.