The premise is everywhere: a flash of genuine emotion crosses someone’s face for a fraction of a second before they can suppress it, and a trained observer can catch it and know they’re lying. It’s the entire concept behind the TV show Lie to Me, a staple of true-crime YouTube commentary, and a claim repeated constantly in body-language content online. It’s also, as an actual practical lie-detection tool, considerably weaker than the popular version of the idea suggests — and the US government spent roughly $900 million finding that out the hard way.
This is a companion piece to reading body language reliably is harder than the videos make it look, focused specifically on the claim that fast, involuntary facial expressions can reliably expose a lie.
The real research behind the idea
Micro-expressions are a genuine, well-documented phenomenon — brief, involuntary facial movements, often lasting a fraction of a second, that can reflect an emotion someone is trying to conceal. The foundational research comes from psychologist Paul Ekman, building on a 1969 paper he co-authored with Wallace Friesen on “nonverbal leakage” — the idea that concealed emotion tends to “leak” through brief, hard-to-control physical signals even when someone is successfully controlling their words. Ekman later claimed, in his book Telling Lies, that in laboratory experiments, truth and lies could be distinguished by facial expression alone with more than 80% accuracy — a figure that became the foundation for an entire industry of lie-detection training built on his work.
Where the claim runs into trouble
The 80%-accuracy figure has not held up well under independent scrutiny. Ekman’s own influential studies have been criticized for selective reporting of methodology, and more importantly, later research specifically testing whether micro-expressions themselves improve lie-detection accuracy found something considerably less impressive: work by psychologist Stephen Porter and colleagues found that untrained observers’ ability to detect deception was only marginally better than random chance — and, critically, that accuracy wasn’t meaningfully improved by the presence of micro-expressions in the people being observed. In other words, the specific mechanism the whole theory depends on — a flash of true emotion that gives away a lie — didn’t reliably help people spot lies any better in controlled testing.
There’s also a more basic technical problem: subsequent research has found that genuine micro-expressions tend to be more subtle than Ekman’s original framework suggested, frequently appearing only in part of the face — the upper or lower half, not both together — which makes them considerably harder to catch reliably in real time than the “watch for the flash of true emotion” framing implies, even for a genuinely well-trained observer.
The $900 million real-world test
The clearest evidence that this doesn’t translate reliably into practice isn’t a lab study — it’s a real, large-scale government program. Starting in 2007, the Transportation Security Administration deployed a behavior-detection program called SPOT (Screening of Passengers by Observation Techniques), built substantially on the premise that trained officers could spot deceptive intent through facial expressions and body language. By the time the Government Accountability Office reviewed it in 2013, the program had cost roughly $900 million.
The GAO’s findings were unusually blunt for a government audit. In 2013, it concluded that ten years after the program’s development, the TSA “cannot demonstrate the effectiveness” of its behavior-detection activities, and recommended limiting future funding. A follow-up 2017 review went further: TSA lacked valid scientific evidence to support 28 of the 36 behavioral indicators it was actually using to flag travelers, and of 178 sources the agency provided to justify its methodology, the GAO found only 3 offered sufficient scientific evidence to support the claims being made. This wasn’t a training or execution failure — it was a foundational problem with the premise the entire program had been built on for a decade, at real cost and with real consequences for travelers flagged based on indicators that couldn’t be scientifically substantiated.
What’s actually real versus what got oversold
The honest version sits in the middle. Micro-expressions themselves are a real, documented phenomenon — brief, sometimes involuntary facial movements do occur, and Ekman’s foundational research on facial expression and emotion (including the widely-used Facial Action Coding System he developed for cataloguing specific facial muscle movements) is genuinely influential and used in legitimate psychological and even animation research today. What didn’t hold up is the specific, practical extension of that research into a reliable, learnable lie-detection skill — the idea that a person, or even a well-trained professional, can watch someone’s face and reliably catch deception through fleeting expressions alone. The gap between “this is a real, interesting facial phenomenon” and “this reliably tells you when someone is lying” is exactly where the TSA lost $900 million, and exactly where most viral body-language content quietly skips a step.
The actual takeaway
Micro-expressions are real. Using them as a practical, trainable lie-detection method is where the science becomes much shakier than popular culture suggests — untrained (and, per the research, often trained) observers perform close to chance, the underlying facial signals are more partial and harder to catch than the original framework implied, and the single largest real-world test of the idea, run by the US federal government for a decade at nearly a billion dollars, couldn’t scientifically justify most of what it was actually looking for. That’s not a reason to dismiss the underlying research — it’s a reason to be skeptical of anyone selling you a trainable shortcut to reliably reading concealed emotion off someone’s face in real time.
For the other major (and similarly overstated) lie-detection method, see the polygraph: what it actually measures, and why courts don’t trust it.
