10.2 — The Face: What It Carries and What It Does Not

In the 1960s Paul Ekman took photographs of posed facial expressions to the Fore people of Papua New Guinea, a group with almost no contact with the outside world. He told short stories — a man's child has died, a man is about to fight, a man sees something that smells bad — and asked people to choose the matching face.

They chose the same faces Americans chose, at rates far above chance.

This became one of the most famous results in psychology: six basic emotions with universal facial expressions — happiness, sadness, anger, fear, disgust, surprise.

It is also, forty years later, one of the field's most seriously contested findings. Both halves matter, and knowing the argument is what lets you read faces accurately.

What the challenge found

Lisa Feldman Barrett and colleagues re-examined the method and found a problem: the original studies gave participants a small list of words to choose from.

When the task is changed to free labelling — describe what this person is feeling, in your own words — agreement drops sharply.

And when the studies were repeated with the Himba of Namibia, another group with limited outside contact, the results looked substantially different from the American pattern when the forced-choice format was removed. People sorted the faces by what the person appeared to be doing — smiling, looking — rather than into emotion categories.

A large review led by Barrett in 2019 concluded that the evidence does not support reliably inferring a specific emotion from a facial configuration. Its central finding: people scowl when angry only some of the time, and scowl when not angry a great deal of the time.

Ekman and others dispute this, and the argument is live. This page takes the position that both sides now largely accept: there is real, above-chance cross-cultural agreement about some expressions, and it is far weaker and far more context-dependent than the popular version claims.

What the face actually carries

Sorted by how much you can rely on it.

Reliable: the presence of strong emotion. That something significant is happening is usually detectable, even when you cannot say what.

Reliable: state changes in people you know. You can detect that your partner's face has changed and be right, because you have thousands of observations of their baseline. This is pattern recognition on a specific person, and it is the most accurate face reading anybody does.

Moderately reliable: broad categories in context. Positive or negative, high or low arousal — 4.1's two dimensions. The face carries the axes better than it carries the labels.

Unreliable: specific emotion from the face alone. Is that anger or concentration? Fear or surprise? Disgust or disapproval? Without context, frequently wrong.

And unreliable: character. Covered in 10.1 and worth repeating because it is the most confident error people make.

Context does most of the work

The finding that reframes the whole subject.

Take a face from a photograph of an emotional moment, put it on a different body in a different situation, and people read a completely different emotion.

Studies doing exactly this — swapping faces between contexts — find that the body and the situation dominate the interpretation, and people are unaware that this is happening. They believe they are reading the face.

The classic demonstration involves the faces of tennis players immediately after winning or losing a point. Shown the face alone, people cannot tell winning from losing. Shown the body, they can immediately.

So when you are confident you read someone's face, you were substantially reading the situation — and where the situation is ambiguous, your reading is much weaker than it feels.

The one signal with decent support

Among all the facial claims, one holds up better than the rest.

A genuine smile involves the muscle around the eye — orbicularis oculi — as well as the mouth. A polite or performed smile frequently uses the mouth alone. The genuine version is called the Duchenne smile, after the nineteenth-century physiologist who mapped it by applying electrical current to a man's face.

What is supported: the eye involvement is associated with genuine positive feeling, and it is harder to produce voluntarily than the mouth movement.

What is overstated: that it cannot be faked at all. Some people can produce it deliberately, and the eye crinkling also appears with intense negative emotion and simply with a wide mouth movement.

Useful as weak evidence. Not a lie detector.

Micro-expressions, honestly

The claim: brief involuntary expressions lasting a fraction of a second reveal concealed emotion, and trained observers can detect them.

What is real: very brief expressions do occur.

What the evidence does not support: that they reliably indicate deception, that they occur frequently enough to be useful, or that training people to spot them improves lie detection to a practically useful degree. Studies looking for them in real deceptive interactions find them rarely.

And there is a large practical test case. The United States Transportation Security Administration ran a behaviour-detection programme built partly on this idea, spending an enormous sum over years. Government audits concluded there was no evidence it worked, and reviews of the underlying science found the behavioural indicators had no established validity for detecting threats.

Treat commercial claims about micro-expression training with the scepticism you would apply to any product promising a superpower for a fee.

Eye contact, which varies more than people think

What is true: direction of gaze tells you where attention is, and mutual gaze is involved in connection and in dominance.

What is culturally variable, substantially: how much eye contact is appropriate, and what avoiding it means. In many South Asian, East Asian and African contexts, sustained direct eye contact with an elder or a superior signals disrespect rather than confidence, and looking down signals respect rather than evasion.

Which makes "he wouldn't look me in the eye" one of the most misused observations there is. It can indicate discomfort, shame, respect, autism, anxiety, cultural norm, or nothing. It does not indicate lying — and there is evidence that liars, aware of the stereotype, sometimes make more eye contact than truth-tellers.

And it varies neurologically. Many autistic people find eye contact aversive or effortful, and require it to be reduced in order to listen properly. Reading that as evasiveness is both wrong and unkind.

Resting faces

Worth a paragraph because it causes real trouble.

Some faces at rest look angry, sad, disapproving or bored. This is structure — the set of the mouth, the brow, the eyelids — and it says nothing about the person's state.

Two consequences.

You are misreading people whose neutral face is not neutral-looking. Almost everybody does this and it affects hiring, first impressions and how people are treated.

And if you have such a face, you are being misread constantly. People conclude you are unhappy, unfriendly or unimpressed when you are simply present. The practical answer is not to change your face but to say the thing out loud — "I'm interested, I just look like this when I'm concentrating" — which sounds absurd and works.

What to do with this

Read the face for state, not for character. State reading in people you know is your best channel; character reading is your worst.

Always take the context first. What is happening, what has just happened, what is at stake. Then read the face against that, rather than reading the face and inferring the situation.

Ask instead of concluding. "You look worried — are you?" One sentence, and it converts a guess into information. People are startlingly willing to tell you, and the guess was going to be wrong a substantial fraction of the time.

Learn one person's baseline properly (10.11). What their neutral face looks like, what their tired face looks like, what their genuinely-fine looks like. This is worth more than any amount of general expression training.

And hold your face readings loosely, particularly the confident ones. Confidence here is produced by fluency rather than by accuracy (10.1), and the reading that arrives instantly and completely is the one most likely to be structure rather than signal.

Next: 10.3 does the same job for the body, including the origin of the statistic that says words are only seven per cent of communication — which is one of the most widely repeated misreadings of a study in existence.