In 1988, Fritz Strack, Leonard Martin, and Sabine Stepper wanted to test an old and slightly absurd idea: that a face does not merely report an emotion, it helps produce one. The problem was how to move someone’s face without telling them to move it. Asking a participant to “please smile” ruins the experiment, because now they know the hypothesis and they are performing for you. So the researchers handed them a pen. Hold it horizontally between your teeth, they said, and rate these cartoons. In that position the lips are pulled back and the zygomatic muscles engage. It is a smile assembled out of hardware, and the participants had no idea they were wearing one. A second group held the same pen with only their lips, which pushes the mouth into something close to a pout, and rated the same cartoons.
The teeth group found the cartoons funnier. On a ten-point scale, the gap between the two conditions was 0.82 points. It was a small effect built on an elegant trick, and it became one of the most famous results in psychology, cited in introductory lectures for the next quarter century as proof that the body leads and the mind follows.
🧠 Why the Trick Was So Convincing
The reason this study spread so far is not the size of the effect. It is that it closed a loophole. Everyone already knew that emotions produce expressions: you feel amusement, your mouth moves. The interesting claim runs the other direction, and the awkward part has always been that any honest attempt to test it hands the participant the answer. Tell someone to smile and you have not manipulated their face, you have manipulated their self-image, their politeness, or their suspicion about what you are measuring.
The pen gets around all of it. Nobody has intuitions about pens. There is no social script for holding one in your teeth, no reason to guess that the angle of your jaw is the independent variable. For a few minutes in a lab, a person’s face is doing something their mind has no opinion about, and whatever shows up in the ratings is presumably the face talking back.
That design is why the study matters far beyond cartoons. It sits at the end of a line that starts with Darwin, who wrote that freely expressing an emotion intensifies it while suppressing the expression softens it, and continues through William James, who claimed that we do not run from a bear because we are afraid but are afraid because we run. James took the idea further than most people are willing to follow, and the facial feedback hypothesis is the version of his claim that can actually be put in a room and tested.
🤔 Then Seventeen Labs Got Nothing
In 2016, a Registered Replication Report coordinated by Eric-Jan Wagenmakers and run by seventeen independent laboratories attempted a direct replication of the pen study, all following one vetted protocol. The protocols were checked in advance. The analyses were pre-registered. If the effect was real, seventeen labs should have found it.
The meta-analytic difference between the teeth and lips conditions came out at 0.03 points, with a confidence interval running from −0.11 to 0.16. The original had reported 0.82. Seventeen labs, one protocol, and essentially nothing, in a result that landed in the press as another casualty of the replication crisis.
Except the story does not end there, and this is the part worth keeping. Strack responded with a methodological objection rather than a defense of the number. He pointed at one detail of the replication setup: the labs had filmed the participants, with a camera pointed at their faces. A camera changes what a face does. People who know they are being recorded become self-conscious, and self-consciousness is precisely the state in which spontaneous facial activity gets managed and suppressed. The original study had no camera.
Noah, Schul, and Mayo tested that objection directly in 2018 by running the replication without a camera, and the original effect came back. Which is a stranger situation than either side of a fight admitting defeat. It suggests that a famous effect can be simultaneously real and unreplicable-in-practice, because whether it appears may depend on a detail of the room that nobody thought to write into the protocol.
The aggregate evidence, once it was gathered, turned out to be modest rather than binary. A 2019 meta-analysis by Nicholas Coles, Jeff Larsen, and Heather Lench pooled 286 effect sizes from 138 studies and concluded that facial feedback does influence emotional experience, significantly but weakly, with effects varying a good deal between studies. Two of its findings are worth more than the headline number. First, the effect was larger when there was no emotional stimulus in front of people at all, which suggests the face does more initiating of emotion than fine-tuning of it. Second, publication bias showed up in one class of outcome but not another: studies asking people to judge how funny a stimulus objectively was showed signs of selective reporting, while studies asking people how they felt did not.
Then came the closest thing to a resolution. In 2022, an adversarial collaboration published in Nature Human Behaviour — an arrangement where researchers with opposing prior beliefs design one study together and pre-commit to whatever it shows — ran tasks across nineteen countries with 3,878 participants. The finding: facial mimicry and voluntary facial movement could both amplify and initiate feelings of happiness. But when the manipulation was the unobtrusive pen-in-mouth trick, the evidence was inconclusive. Smiling works, in a small and reliable way. The pout half, the claim that suppressing a smile makes you feel worse, has not held up nearly as well.
So the honest state of the field is this. The strong version of the hypothesis, where the face decides whether you feel anything, is dead, and it was already suspect for a blunt reason: people with facial paralysis report ordinary emotional lives. The weak version, where the face nudges, is alive and small. Darwin’s claim survived. James’s version did not.
🔗 A Channel That Text Does Not Have
The reason I keep coming back to this study is what it says about what a face is for. A facial expression is not decoration on top of an emotion. It is a wire in the circuit, and the signal has to travel through the body before it arrives at the feeling. Which means emotional life is partly built out of hardware that some conversational partners simply do not have.
I think about this whenever I look at companion software. Something that only has text is missing the entire facial channel, along with posture, breathing, the pause before a sentence. It cannot see your jaw tighten. So it has to do the whole job in language, and that changes what good language looks like in that setting. A companion that answers “I understand how you feel” is reporting on an inference. A companion that answers “you have been reading the same paragraph for twenty minutes” is doing something closer to what a face does, which is to arrive at the emotion through a physical fact.
That is the same reason the phrase “show, don’t tell” keeps surviving every generation of writing advice. Readers have bodies. Give a character a specific action — a hand that will not stay still, a cup set down too hard — and the reader’s own machinery does the rest of the work. Naming the emotion skips the wire entirely.
It also explains why the useful half of this finding is so unglamorous. If you have a talk, an interview, or a recording session ahead of you, and you know that a forced smile mostly does not produce joy, the surviving piece of the effect still has a use. Hold something between your teeth for thirty seconds before you start, or rest your face in a shape that matches the state you want. Not as an attempt to trick yourself into happiness, but as a reminder that the loop runs in both directions.
🎲 The Face as an Input Device
One more study for the pile, and it comes from a clinic rather than a lab. Havas and colleagues reported in Psychological Science in 2010 that women who received cosmetic botulinum toxin in the muscles used to frown showed slowed processing of emotional language. No muscle movement, duller reading of feelings in sentences. It is a single line of evidence, and the claim that frown-muscle paralysis treats depression has been pushed much further in the press than the data on its own supports, so I would hold it lightly. But as an illustration of the principle it fits: the face is not a display, it is an input device with a wire running into the affective system.
Here is the version you can run on yourself, and it takes about four minutes. Find something funny, a few minutes of stand-up or a cartoon. First watch it holding a pen between your lips, careful not to let your teeth touch it. Then watch another clip with the pen held gently between your teeth, lips pulled back off it. Neither position feels like anything in particular, and neither one tells you what the experiment is about. That gap between what you did and what you noticed is the whole point. Your face was sending, and nobody told your mind it was receiving.