Reading the Eyes test: an open online version with new photographs

Each screen shows a black and white strip of one person's eyes, with four words around it. Pick the word that best says what the person is thinking or feeling. 40 strips, a practice one before them, and no clock running.

Start Reading the Eyes

40 photos, 4 words each · about 7 min · no time limit, phone or computer

At the end you see how many intended words you chose, and a free account keeps that count.

In brief

Reading the Eyes is a social cognition test. A photograph of the eyes and four words: pick the one that says what the person is thinking or feeling. It is free to take: the first test needs no account, later ones a free account. It takes about 7 minutes and is available in English and Spanish. The result screen shows your own scores; this version does not turn them into a percentile from published norms. A test, not a diagnosis.

Which word fits these eyes?4 words · 1 intended · take your time

How it works

One practice photo, then 40 that count, in the same fixed order for everyone. Nothing is timed against you, and most people are done in about 7 minutes.

  1. 1 · First

    1 practice photo

    The first photo carries a Practice label so you can learn the layout before anything counts. It is not scored and tells you nothing about your answer. A short line then says the real photos begin.

  2. 2 · Each photo

    Four words around the eyes

    Two words sit above the strip and two below. Tap or click the one that fits best. With a keyboard, keys 1 to 4 choose top left, top right, bottom left and bottom right. The words swap places on every photo, so the intended one has no fixed spot.

  3. 3 · Any time

    What the words mean

    Under each strip, a button called What do these words mean? opens a one-line meaning for each of its four words. Open it as often as you like; it never touches the count. Taken on its own, the task also lists every word with its meaning before the practice.

  4. 4 · At the end

    A count, not a band

    The last screen says on how many of the 40 photos you chose the intended word, for example 27 of 40. No band or percentile follows it, because the group to compare with is still being built.

There is no feedback, no going back and no time limit. As soon as you pick a word, the next photo appears. For each photo the task keeps your word, where it sat on the screen and how long you looked, but only the word decides the count.

The format comes from Simon Baron-Cohen and colleagues at Cambridge. Their revised version, published in 2001, showed 36 black and white photographs of the eye region with four words each, and came with a glossary for anyone unsure of a word. This version is modelled on the Reading the Mind in the Eyes format, but every photograph and every word set is new, made for this site in English and Spanish, with 40 photos rather than 36.

The faces are 20 women and 20 men aged 18 to 75, 12 of them 55 or older, drawn from 8 groups of ethnic background with 5 faces each; 2 wear glasses. On 18 photos the person looks straight at you and on 22 the gaze turns away, spread across positive, negative and neutral states, so the direction of the gaze alone never names the answer.

What it measures

The task asks you to read a state of mind from a small, still part of a face. Psychologists call this mental state recognition, and it is one part of social understanding.

What it draws on

  • Reading the brows, the lids and the skin around the eyes, with no mouth, voice or posture to help.
  • Linking what you see to a word for a thought or a feeling, such as doubtful, curious or proud.
  • Knowing emotion words in the language you take it in, which is why every word has a meaning one tap away.

What it leaves out

  • Movement, tone of voice, context and the back and forth of a real conversation, which carry most social signals in daily life.
  • How you get on with people you know, or how tiring social situations feel to you.
  • Speed: response times are kept with the data, but the count ignores them.

In the 1997 study that introduced the format, adults with autism or Asperger syndrome chose the intended word less often than two comparison groups, while they did as well as others at telling the sex of a person from the same eye region. The 2001 revision again separated 15 adults with Asperger syndrome or high-functioning autism from 239 adults without, and in both groups scores fell as self-rated autistic traits rose. These are group averages; many people in each group scored alike.

A count here also moves with vocabulary, with the language you take it in and with how rested you are. It describes one sitting, not a fixed quality of a person.

A test, not a diagnosis.

This task is one part of social understanding. It is not an autism test on its own, and no count on it can diagnose or rule out anything.

What you get

The count is on screen as soon as you answer the 40th photo.

On the result screen

  • On how many of the 40 photos you chose the intended word, as a plain count.
  • A line saying the comparison group is still being built, so there is no band and no percentile yet.
  • A reminder that this is one task and not a diagnosis, and a Continue button.

After that

  • A free account keeps the count, so you can open it from any device.
  • The same task opens the autism traits check, which shows it next to 42 statements about social situations, routines and the senses.
  • One row per photo is saved with the result: the word you picked, the intended word, its position and your time. These rows feed the calibration described below.

How this version is scored

Each photo is stored with its id, the word you picked, the intended word, whether the two match, the position of your pick and your time in milliseconds.

Intended word
The word the photo was made for. The other three share its tone, positive, negative or neutral, so mood alone never gives it away, and every intended word turns up as a wrong option on some other photo.
Count
The number of photos out of 40 where your pick was the intended word. The screen shows it as a count, and the saved data also holds it as a percentage.
Practice
The practice photo is stored apart from the 40 and never counts.
Time
Measured from the moment a photo has loaded to your pick. It is kept so the photos can be studied, and has no part in the count.
Keeping a photo
Once a photo has enough answers, it stays only if at least half of people pick the intended word and no other word draws more than a quarter. This is the rule the 2001 revision used. A photo that fails is retired, and the total becomes the number of photos kept.

Why there is no band yet. No published table fits new photographs and new words, so the screen shows your count only. Bands will come from people who take this version, once each photo has 225 completed answers, and they will say plainly that the comparison group is our own.

Open material. No photograph shows a real person: every image was generated for this task, then picked by hand from several takes against a written description of the brows and lids. The images are released under CC0, and the words, their meanings and the item data under CC BY 4.0. The method page listed under Sources has the full method and the download.

Before you start

Without a clock, the setting matters less than in a timed task. The language and the screen matter more.

For a cleaner attempt

  • Set aside about 7 quiet minutes and use the language you know best.
  • Pick a screen big enough to see brows and lids clearly, and turn the brightness up.
  • Open the meanings whenever a word is unclear; that is what they are there for.
  • Take the time you need on each photo. There is no skip, so every photo gets one word.

Interpret cautiously when

  • The task is not in your first language, or some emotion words were new to you.
  • Low vision, a small or dim screen, or glare made the eyes hard to see.
  • You were tired or distracted, or tapped quickly just to reach the end.
  • Faces from some backgrounds are less familiar to you; the set spreads its faces across 8 groups to soften that effect.

Questions

Is this the Cambridge Reading the Mind in the Eyes test?

No. It is modelled on the format Baron-Cohen and colleagues published in 2001, but the photographs and the words were made new for this site. None of the original pictures or word sets are reused.

Are the people in the photos real?

No. Every image was generated by a computer and shows nobody who exists. That keeps real faces out of the task and lets the images be released under CC0, so anyone can reuse them.

What is a good score?

There is no comparison group yet, so the page does not call any count high or low. Once enough people have answered each photo, counts will be compared with people who took this same version. Until then the count tells you how many intended words you chose, and nothing more.

Is Reading the Eyes an autism test?

No. It looks at one part of social understanding, reading a state of mind from the eyes. Autism is diagnosed by a qualified clinician after a full assessment. The autism traits check on this site starts with this task and adds 42 statements, and it cannot diagnose autism either.

Can I take it on my phone?

Yes. The four words sit around the photo within reach of a thumb, and nothing is timed. Turn the brightness up so the brows and lids are easy to see.

Can I take it again?

Yes, but the photos come in the same order every time and only the word positions change. On a second attempt you still remember the first, so it is not a fresh measure.

Is the Reading the Eyes test free?

Yes. The Reading the Eyes test is free to take: your first test needs no account, and later tests need a free account so your results are kept.

How long does the Reading the Eyes test take?

About 7 minutes. Your result appears on screen as soon as the test ends.

Sources

By Cristian Dominguez Rein-Loring, Creative Technology Lead · Updated October 25, 2026