One-tenth of a second is shorter than a blink. In a series of Princeton experiments, it was still enough time for people to decide whether an unfamiliar face looked trustworthy, competent, likeable, attractive or aggressive.
Their 100-millisecond judgments closely resembled the ratings made by another group who could inspect the same faces without a time limit. Giving viewers half a second or a full second did not significantly improve that agreement. Extra time did something else: it tended to increase confidence in impressions that had already taken shape.
This is a finding about the speed and stability of face-based impressions, not their truth. The researchers did not determine whether the photographed people were genuinely trustworthy or competent. Agreement with other observers is not the same thing as accuracy about a stranger’s character.
How the researchers compressed a first impression
Janine Willis and Alexander Todorov recruited 245 Princeton undergraduates for the 2006 Psychological Science study. A preliminary group of 128 rated the faces without time limits. Their average ratings became the criterion against which rapid judgments were compared.
The remaining 117 students participated across five experiments. Each experiment asked about one trait: attractiveness, likeability, competence, trustworthiness or aggressiveness. The trait-specific groups were small, ranging from 20 participants for attractiveness to 25 for likeability and aggressiveness. There were 23 in the competence experiment and 24 in the trustworthiness experiment.
The stimuli were 66 frontal headshots of unfamiliar amateur actors, evenly divided by sex and drawn from a standardised face database. The actors were aged 20 to 30, wore grey T-shirts and held neutral expressions. The photographs excluded visible makeup, glasses, earrings and facial hair. Each face appeared for 100, 500 or 1,000 milliseconds, after which participants made the trait judgment and rated their confidence.
A tenth of a second already reproduced the pattern
At 100 milliseconds, the pattern of rapid ratings across faces was already highly correlated with the unrestricted group’s pattern. The zero-order correlation was 0.73 for trustworthiness and 0.52 for competence. It was 0.59 for likeability, 0.52 for aggressiveness and 0.69 for attractiveness.
Longer exposure did not significantly increase these correlations for any of the five traits. That is the basis for the claim that more viewing time did surprisingly little to change the impression. The Association for Psychological Science’s contemporary summary described the core result as a first impression formed in a tenth of a second and not substantially altered by longer looks.
The statistic needs careful wording. A correlation across average ratings does not mean every participant gave the same answer at every duration. It means faces that tended to receive higher ratings without a deadline also tended to receive higher ratings after the brief exposure. The group-level ordering was already visible.
Extra time strengthened confidence more than judgment
Moving from 100 to 500 milliseconds was not completely inert. Participants’ ratings became somewhat more negative, their responses after the image disappeared became faster and their confidence increased. From 500 to 1,000 milliseconds, most trait judgments changed little, while confidence increased for four of the five judgments.
Confidence was recorded on a seven-point scale after each decision. The result does not mean that first impressions can never be revised. The experiment compared fractions of a second spent viewing the same static photograph; it did not expose participants to contradictory behaviour or personal history. Its narrower implication is that repeated visual sampling can create a subjective feeling of certainty before evidence relevant to character has appeared. A longer look and a better basis for judgment are not necessarily the same thing.
The researchers also found that impressions became more differentiated with time. At 100 and 500 milliseconds, ratings largely collapsed onto a broad positive-versus-negative dimension. By one second, aggressiveness and trustworthiness formed a second dimension separate from the cluster of positive traits. More time added some nuance even though it did not meaningfully improve agreement with unrestricted ratings.
A 2009 follow-up on trustworthiness pushed the timing lower. Agreement with unrestricted judgments rose above chance after 33 milliseconds, improved with additional exposure and stopped improving beyond 167 milliseconds. Together, the studies suggest that an appearance-based evaluation begins extremely quickly and then becomes anchored.
Consensus about a face is not knowledge of a person
The criterion ratings came from people viewing the same photographs without a time constraint. They were not reports from friends, behavioural tests or records of how the actors treated others. The 0.73 trustworthiness correlation therefore measures agreement between rapid and unhurried observers. It does not show 73 per cent accuracy at detecting who deserves trust.
Todorov has since emphasised the danger of confusing the two. A Princeton profile of his later work describes facial first impressions as compelling but faulty. A 2021 meta-analysis of trustworthiness accuracy found statistically significant but modest relationships between perceived and measured trustworthiness, with a face-level correlation of 0.14. Its authors concluded that the accuracy was unlikely to have practical utility.
Even the photograph is unstable evidence. In later work, Todorov and Jenny Porter found that different images of the same person produced widely different impressions. Lighting, expression, angle and the moment captured can shift ratings enough that the image may tell us as much about the photograph as about the person.
The study was controlled, small and socially narrow
The controlled headshots were useful because they reduced visual noise, allowing the researchers to isolate exposure time. They also limit what the study can represent. The observers were students from one American university, the photographed actors were young adults, and each main experiment contained only 20 to 25 participants.
The faces were also presented repeatedly within a highly structured task. Participants knew they had to make a trait decision and were given the relevant label in advance. That differs from an ordinary encounter in which a person may notice many cues without being explicitly asked to classify a stranger as trustworthy, competent or aggressive. The task reveals how quickly a requested inference can be produced, not how often each inference spontaneously dominates everyday attention.
Real encounters include movement, voice, posture, clothing, setting, group membership and behaviour. Observers also bring cultural experience and stereotypes to the task. The experiment shows that people can rapidly extract a shared impression from this particular class of images. It does not establish that every population interprets every kind of face in exactly the same way.
Nor did the researchers prove that 100 milliseconds is the minimum possible threshold. Their conclusion was that it was sufficient. The later trustworthiness study found evidence of agreement at shorter exposures, illustrating why a tested duration should not be mistaken for a hard biological boundary.
Fast judgments can matter even when they are wrong
The finding is consequential because face-based impressions can enter decisions before stronger evidence arrives. They may influence which applicant appears capable, which defendant looks dangerous or which candidate seems fit to lead. ScienceBlog has reported how facial stereotypes can affect sentencing judgments, a setting in which unsupported confidence carries an obvious cost.
The useful lesson is not that people should obey their first impression. It is that deliberation may begin after the impression has already formed. Simply staring longer at the same face may add confidence without adding relevant information.
A better correction is new evidence: behaviour over time, context, conversation and records connected to the trait being judged. The Princeton study revealed how quickly the mind constructs a social verdict from appearance. Later work supplies the warning that the speed and confidence of that verdict should never be confused with access to character.
