The technique is simple to describe and difficult to do. You take film of a person's face, and you slow it down until you can see individual frames. Things appear that were not visible at normal speed.
Ekman and Friesen called them microexpressions: complete facial expressions of emotion that appear and disappear in somewhere between a twenty-fifth and a fifth of a second, often across a face that is otherwise composed. In the Mary interview, slowed to frame-by-frame, a flash of anguish crossed her face and was replaced by a smile within a fraction of a second.
The proposed mechanism is a race between two motor pathways. Emotional expressions are generated by subcortical circuits — the amygdala and related structures projecting to the facial nucleus — which are fast and not under voluntary control. Voluntary suppression is cortical, and arrives late. The interval between the involuntary onset and the deliberate override is the microexpression. Neurological evidence supports the two-pathway model: patients with certain motor cortex lesions cannot produce a smile on command but smile normally at a joke, and patients with certain subcortical damage show the reverse dissociation. The pathways are genuinely separate.
The measurement problem is where this gets serious. If you ask an observer what emotion a face is showing, you get an interpretation, and interpretations are contaminated by context, expectation and stereotype. Ekman and Friesen's answer was the Facial Action Coding System, published in 1978: an exhaustive taxonomy of the visible movements the facial musculature can produce, each labeled as an action unit, with no emotional interpretation attached. AU4 is a brow lowerer. AU12 is a lip corner puller. What a coder records is a set of action units and their intensities and timings, and only afterwards is any inference drawn.
This is the discipline that separates forensic observation from intuition, and it is expensive. FACS certification takes roughly a hundred hours. Reliable coding of a short clip can take an hour per minute of video.
That cost is why the commercial version of this material is almost always something else wearing its clothes. Microexpression training marketed as a two-day course, or an app, is not teaching FACS. It is teaching a small set of prototypes, and the transfer of that training to real accuracy is weak. When Ekman's own claims about training-based improvements have been tested independently, the results have been mixed, and the security screening program built on this logic failed comprehensively — chapter 56.
The honest summary of the state of the evidence is narrow. Microexpressions exist and can be reliably coded. Their frequency in real deception is low; most liars do not produce one, so their absence means nothing at all. When they do occur, they indicate a concealed emotional state — which, as the previous chapter argues, is not the same as a lie. Some experimental work, including a 2018 study on lies about future intentions, finds discriminative value; the effect sizes are modest and the laboratory conditions generous.
Read as an inconsistency detector rather than a lie detector, the tool is real. A flash of contempt across a face that is agreeing with you is information about the relationship. It is not information about the truth of the sentence.
See Defense: verifying tells against baseline