Micro-Expression Research Methods
Studying a brief facial movement requires a timing rule, a coding method, and a cautious interpretation. None of those steps turns a recorded face into a reliable account of someone's private thoughts.
Published and updated November 1, 2026.
Origins in close review of filmed behavior
Haggard and Isaacs examined filmed psychotherapy sessions with slowed playback in their 1966 work. Their chapter concerned micromomentary facial expressions in psychotherapy. That historical setting should not be silently converted into a modern claim about detecting lies.
Ekman and Friesen's 1969 paper addressed nonverbal leakage and clues to deception. Its title identifies an important research topic, not proof that a contemporary tool detects deception. Keep historical references separate from any performance claim about a new system.
When introducing a project, explain whether it studies timing, observable movement, or an interpretation framework. Those are related questions, but they require different evidence. A clear research question prevents a literature summary from becoming an unsupported promise.
Define micro and macro before collecting labels
Paul Ekman Group describes macro expressions as lasting about one-half second to four seconds. It describes micro expressions as one-half second or less, sometimes as short as one twenty-fifth second. Definitions vary between researchers, so name the convention used in your study.
State how onset and offset are selected and how uncertain boundaries are handled. Explain whether reviewers classify an entire event or only an interval within a longer movement. Do not present a duration category as proof of a concealed emotional state.
Inspect the recording and playback information available to your team before making timing claims. If timing cannot be established adequately, document that limitation. Avoid adding unsupported universal frame-rate requirements merely because a movement is described as brief.
Design stimuli and recording around the question
Write down what participants will be asked to do and why the footage is needed. Explain recording, access, retention, and possible reporting before seeking appropriate informed consent. Keep participation voluntary and avoid instructions that promise to uncover hidden feelings.
Consider lighting, camera angle, occlusion, and whether the face remains visible during the task. These are practical inspection questions rather than numerical capture standards. Record changes in conditions so reviewers can distinguish visibility problems from movement differences.
Use FACS as a movement vocabulary
Paul Ekman Group describes FACS as an anatomically based system for visually discernible facial movement. Its Action Units describe components of movement instead of verifying someone's emotion. The system was first published by Ekman and Friesen in 1978 and revised in 2002.
The FACS overview describes intensity from A, trace, to E, maximum, alongside side and asymmetry notation. Choose only the distinctions your reviewers are trained to apply. A short project codebook should not claim to replace the manual or establish proficiency.
Separate movement labels from any interpretation assigned later in the analysis. Include options for uncertainty, insufficient visibility, and events outside the selected coding scope. Keep action descriptors distinct from AUs when your annotation scheme uses both.
Establish independent coding and agreement
The FACS overview recommends agreement between at least two independent coders. Give reviewers the same definitions and practice material, then ask them to annotate independently. Discuss disagreements only after preserving their initial coding decisions.
Clarify whether agreement concerns event presence, intensity, duration category, or exact time boundaries. Different questions should not be collapsed into an unexplained reliability score. Update ambiguous codebook instructions before using separate material to check the revised procedure.
As of November 1, 2026, Noldus lists Cohen's kappa within The Observer XT's reliability features. That published option can inform tool selection, but it does not choose an appropriate analysis for you. Explain the agreement measure, unit of coding, and treatment of missing observations.
Separate spotting from interpretation
Finding a short interval of movement is not the same task as explaining its meaning. Record what a spotting procedure marks before attaching expression names. Inspect false alarms and missed intervals alongside examples that appear to work.
Use held-out clips when changing thresholds, coding rules, or a model-assisted review process. Do not tune a method on every example and describe the same collection as independent evidence. Keep the evaluation description honest about which decisions were made after inspecting the material.
Open tools and licensing boundaries
As of November 1, 2026, OpenFace describes landmarks, head pose, Action Unit recognition, and eye gaze. Its README directs commercial users to a separate commercial license through CMU. Do not assume an accessible repository means every intended use has the same license terms.
As of November 1, 2026, Py-Feat describes AU, emotion, and landmark detection from images and video. Its code is MIT-licensed, but pretrained weights have separate terms. Version one uses sixty-eight-point landmarks, while version two adds a MediaPipe mesh and other outputs.
These published capabilities do not establish a complete, validated micro-expression spotting pipeline. Treat the tools as possible components and define how any proposed spotting step will be tested. Record versions, configurations, annotation rules, and license checks in the project documentation.
Keep annotation tools distinct from automatic estimates
As of November 1, 2026, BORIS provides free, open-source event logging for video, audio, and live observation. Its GPL v3 description lists Linux, Windows, and macOS, and its scholarly reference is Friard and Gamba, 2016. An event log remains a record of the coding decisions entered into it.
Choose a review tool by the timing and annotation structure your question requires. Test exports and clip references before processing the entire collection. Keep manual annotations distinguishable from imported model estimates so later reviewers know where each label originated.
Present the interpretation debate directly
Paul Ekman Group's position is that micro expressions can reveal concealed emotion. Attribute that position explicitly rather than presenting it as an uncontested scientific conclusion. It does not authorize claims that micro-expressions establish truthfulness or reliably detect lies.
Barrett and colleagues' 2019 review describes substantial variation across cultures, situations, and people. A facial configuration can accompany different emotions or no emotion. This challenges a simple mapping from a visible pattern to one verified internal state.
Jack and colleagues reported differences between Western and East Asian representations of basic expressions in 2012. A reply argued that commonalities outweighed those differences. Explain the active debate without using uncertainty as a reason to abandon careful movement description.
Ethics, law, and reporting
EU AI Act Article 5(1)(f) prohibits workplace and educational emotion inference, with medical or safety exceptions. The prohibitions apply from February 2, 2025. Talk to counsel about your use case. This guide is not legal advice.
Do not use a research label to justify hiring, promotion, diagnosis, threat assessment, student scoring, or surveillance. Decide who can inspect recordings and how participants' rights will be handled. Report only what the consented purpose and available evidence support.
Include duration definitions, coder training, agreement procedures, recording limitations, and unresolved disagreements in the methods record. Describe any software as an estimator, not an independent witness to private experience. Avoid accuracy claims unless an appropriate completed evaluation actually supports them.
A practical research checklist
Confirm the question, consent process, duration convention, visibility rules, and coding scope before annotation. Preserve independent labels and source intervals during disagreement review. Keep held-out material separate from clips used to refine the method.
At reporting time, distinguish observed movement from interpretation and model-generated categories. State which definitions varied and which limitations remain unresolved. A careful study can describe brief visible events without claiming access to thoughts, feelings, or deception.