Paul Ekman, Micro-Expressions, and the Science of a Real Smile

The scientific understanding of what makes a smile genuine didn't arrive all at once. It was built across 150 years by researchers who approached faces methodically — measuring muscles, coding movements, testing cross-cultural recognition. This post covers that history, the specific findings that matter for everyday smiling, and how those findings translate into what a smile coach actually measures.

1862: Duchenne stumbles into it

Guillaume Duchenne (1806–1875) was a French neurologist studying how electrical stimulation could contract facial muscles. His method sounds grim now — he'd apply electrodes to the faces of subjects (including, reportedly, an "old toothless man with a face not much disfigured") to trigger individual muscle contractions, then photograph the result.

In his 1862 book Mécanisme de la physionomie humaine, he observed something specific: a smile produced by electrical stimulation of the zygomatic major looked different from a smile produced by genuine amusement. The electrical version only moved the mouth. The genuine version also contracted the muscle around the eye — the orbicularis oculi — producing what Duchenne called "the sweet emotions of the soul".

He concluded that this muscle was under involuntary control: "the play of this muscle...does not obey the will, it is only put into action by a true feeling, by an agreeable emotion. Its inertia, in smiling, unmasks a false friend." For more on the anatomy, see our smile muscles explained post.

1960s–1970s: Ekman makes it systematic

Duchenne's observation sat mostly dormant until Paul Ekman picked it up a century later. Ekman's early research (late 1960s) was on whether facial expressions of emotion are universal across cultures or learned culturally. Traveling to isolated communities in Papua New Guinea that had limited contact with Western media, he showed photographs of different expressions and asked subjects to identify the emotion.

The answer: the basic expressions — happiness, sadness, anger, fear, surprise, disgust — were recognised across cultures at rates far above chance. The genuine smile specifically was universal. This was strong evidence that facial expressions of emotion are biological rather than cultural.

1978: FACS arrives

With Wallace Friesen, Ekman published the Facial Action Coding System (FACS) in 1978. FACS is a taxonomy: every visible facial muscle movement gets a numbered Action Unit (AU). There are 44 main AUs. A facial expression is coded as a set of AUs with intensities.

For smiles:

• AU12 — lip corner puller (zygomatic major). This is the "mouth smile" muscle.
• AU6 — cheek raiser (orbicularis oculi orbital portion). This is the "eye crinkle" muscle.
• AU25 — lips part (opens the mouth to show teeth).

A polite smile is, in FACS terms, AU12 alone. A genuine Duchenne smile is AU6+AU12 — and that's the specific notation that underlies everything the scientific community means by "genuine smile".

FACS is still the standard. Every serious piece of modern facial expression research uses it. When an AI system analyses smile authenticity, what it's fundamentally doing is estimating AU6 intensity, AU12 intensity, and their relative timing.

Micro-expressions

Parallel to the smile work, Ekman and colleagues identified what they called micro-expressions — involuntary facial movements lasting 1/25th to 1/2nd of a second that briefly express a real emotion even when the person is consciously trying to suppress it.

These are most visible during attempts at deception or emotional masking. If someone is pretending to be pleased but actually isn't, a micro-expression of their real emotion (contempt, disgust, fear) may flash across their face for a fifth of a second before they recompose.

For smile research specifically, micro-expressions matter in two ways:

• A micro-smile can flash across someone's face during an otherwise neutral expression, revealing genuine amusement they didn't express socially
• The absence of AU6 during a sustained AU12 (polite smile) is a visible "leak" that the smile isn't genuine — and observers detect it within about 50 milliseconds, even when they can't articulate why

Modern replications

The core findings — that AU6 activation distinguishes genuine from polite smiles, that this is largely involuntary, that observers can detect the difference — have been replicated extensively. Some specific modern findings:

• Harker & Keltner (2001): analysed college yearbook photos, found that women with genuine (Duchenne) smiles had better marital and personal outcomes 30 years later. The effect was specific to genuine smiles, not general smile presence.
• Abel & Kruger (2010): analysed baseball players' yearbook photos, found that those with genuine smiles outlived those with polite or no smiles by an average of about 7 years.
• Krumhuber et al. (2007): showed that observers consistently rated Duchenne smiles as more trustworthy and authentic than non-Duchenne smiles in controlled video experiments.

The smile-specific science is stable. The extensions to deception detection and real-time emotion-reading are more contested — Ekman's applied work there has drawn criticism — but the smile muscular pattern itself is uncontroversial.

How this applies to smile coaching

Any serious smile-training tool is doing the same thing, at its core: measuring AU6 and AU12 activation, their timing relative to each other, and their symmetry across left and right. That's the full scientific content of "is this a genuine smile?". Everything else — mouth shape aesthetics, teeth position, lip plumpness — is cosmetic commentary, not authenticity measurement.

What modern AI smile coaches (including ours) do in practice:

1. Detect facial landmarks around the eyes and mouth
2. Estimate AU6 and AU12 activation intensity in real time
3. Compute a ratio — is the eye-crinkle proportionate to the mouth lift?
4. Track symmetry between left and right sides
5. Feed back a score that users can train against

The underlying scientific model is 150 years old. The execution is new — on-device computer vision fast enough to give feedback 30 times per second. The combination is what makes smile training possible at a scale no human coach could offer.

The pattern, one more time

Stripped of detail, the 150-year scientific conclusion is this: a smile is genuine when the orbicularis oculi activates alongside the zygomatic major. The first is largely involuntary and signals authentic positive emotion. The second is voluntary and signals "I am presenting a smile". A smile with the first reads as real; a smile with only the second reads as performed. Humans detect the difference in under a tenth of a second. Everything else in smile science is elaboration on that core finding.