FACS and Action Units Explained
FACS provides a vocabulary for describing visible facial movement without assuming what someone feels. Understanding its notation helps reviewers distinguish anatomical descriptions, expression prototypes, and software estimates.
Published and updated November 1, 2026.
Six Action Units and their muscle basis
| Action Unit | Movement | Muscle basis, from the FACS overview |
|---|
| AU1 | Inner brow raiser | Frontalis, pars medialis |
| AU2 | Outer brow raiser | Frontalis, pars lateralis |
| AU4 | Brow lowerer | Depressor glabellae, depressor supercilii, corrugator supercilii |
| AU6 | Cheek raiser | Orbicularis oculi, pars orbitalis |
| AU12 | Lip corner puller | Zygomaticus major |
| AU15 | Lip corner depressor | Depressor anguli oris (triangularis) |
Prototype combinations, not proof of feeling
| Prototype label | Commonly cited EMFACS-style combination |
|---|
| Happiness | 6+12 |
| Sadness | 1+4+15 |
| Surprise | 1+2+5B+26 |
| Fear | 1+2+4+5+7+20+26 |
| Anger | 4+5+7+23 |
| Disgust | 9+15+17 |
| Contempt | R12A+R14A |
What FACS describes
Paul Ekman Group defines FACS as a comprehensive, anatomically based system for visually discernible facial movement. It breaks expressions into individual components of muscle movement called Action Units. An AU label describes a visible action rather than a verified psychological state.
This distinction matters whenever coding results leave the original research notebook. A chart labeled with an emotion word may conceal several separate decisions about movement and interpretation. Keep those decisions explicit so readers can understand what reviewers actually observed.
Action descriptors are separate from Action Units in the FACS overview. Do not treat every notation attached to a face as an AU. Define the vocabulary your project uses before comparing annotations or interpreting software output.
Origins and editions
Paul Ekman Group credits Swedish anatomist Hjortsjö with groundwork in 1970. Ekman and Wallace V. Friesen first published the FACS manual in 1978. The system was revised in 2002, and Wikipedia credits that edition to Ekman, Friesen, and Joseph C. Hager.
These dates describe the coding system's development, not proof that every interpretation attached to it is settled. A movement vocabulary and a theory of emotional expression answer different questions. Use the manual for coding rules and separate research references for claims about interpretation.
When documenting a study, name the edition and explain any simplified coding rules. If a team uses only selected movements, state that scope rather than implying comprehensive coverage. Preserve these choices when comparing a later annotation round with an earlier one.
How Action Units are defined
The muscle table summarizes six examples from the FACS overview rather than reproducing the full manual. AU1 concerns inner brow raising, while AU2 concerns outer brow raising. Both involve the frontalis, but the cited muscle portions differ.
AU4 describes brow lowering and lists several muscles in the overview. AU6 describes cheek raising, while AU12 describes lip corner pulling. AU15 describes lip corner depression and should not be treated as a diagnosis or personal characteristic.
Use these names to guide a precise movement description, not to shortcut training. Check whether the relevant area remains visible throughout the interval being coded. If head movement or occlusion prevents a decision, record uncertainty rather than assigning a convenient category.
Intensity from A to E
The FACS overview describes an intensity scale running from A, trace, to E, maximum. Those letters qualify movement coding rather than directly measuring the strength of an internal feeling. Make that distinction clear in table headings and figure captions.
Write down how your reviewers apply intensity criteria before coding the main collection. Use practice footage to identify disagreements and clarify ambiguous instructions. Keep the original annotations when reaching a consensus so later reviewers can see what changed.
Do not assume a vendor's numerical score is interchangeable with a FACS intensity letter. A software scale needs its own documentation and a task-specific evaluation. A value that looks precise does not, by itself, establish what movement occurred.
Side, unilateral, and asymmetric notation
Wikipedia's overview describes R and L for right and left, and U for unilateral coding. It also describes A for asymmetric notation in the relevant coding context. Because A also appears in intensity notation, explain the notation rather than displaying isolated symbols.
Specify whether side refers to the participant's face or a viewer's screen perspective in your project instructions. This is a practical documentation recommendation, not a replacement for the manual. Apply the same convention across your codebook, annotation exports, and review reports.
Inspect whether cropping, face angle, or obstruction makes one side harder to review. An apparently unilateral movement may require a visibility note rather than an immediate interpretation. Preserve the source clip so another coder can check the same interval.
Read prototype combinations cautiously
The combination table lists commonly cited EMFACS-style prototypes from the FACS overview. The emotion labels identify prototype associations, not a determination of a person's feelings. AU6 with AU12, for example, is listed with the happiness prototype.
A combination can describe a pattern while leaving its meaning unresolved. The participant, situation, timing, and nearby behavior remain relevant to interpretation. Do not turn a prototype table into a checklist for judging honesty, personality, or suitability.
Some combinations include intensity or side notation alongside AU numbers. Preserve the notation when citing the source, and explain how your own project uses it. Avoid silently converting a descriptive prototype into a software performance claim.
EMFACS selects emotion-relevant events
Paul Ekman Group describes EMFACS as applying FACS selectively to emotion-relevant events. It says EMFACS instructions are available to people who have passed the FACS Final Test. Wikipedia credits EMFACS-7 to Friesen and Ekman in 1983.
Selection changes the coding scope, so explain what your team includes and leaves out. A selective scheme should not be presented as a complete record of every facial movement. Describe how reviewers handle events that fall outside the chosen categories.
The term emotion-relevant describes the framework's focus, not certainty about an individual's experience. Keep the broader interpretation debate visible when reporting EMFACS-style patterns. Readers should be able to distinguish the selected coding scheme from conclusions added afterward.
Baby FACS is a separate adaptation
Wikipedia credits Harriet Oster with developing Baby FACS for infants and young children. That attribution does not mean adult coding assumptions can simply be transferred without appropriate training. State which framework your review uses and why it fits the material.
As of November 1, 2026, Noldus lists Baby FaceReader for ages six to twenty-four months, based on Baby FACS. This is a description of that published product offering, not an independent validation claim. It does not establish an infant-analysis capability for Mental Edge.
Training and proficiency
Paul Ekman Group describes a 527-page manual and a 197-page Investigator's Guide. Its published estimate for self-instruction is roughly fifty to one hundred hours. It describes the FACS Final Test as a thirty-four-item video-based proficiency test.
Those details help distinguish learning a few labels from developing coding proficiency. Wikipedia notes that reliable FACS coding usually relies on certified coders. It recommends agreement between at least two independent coders rather than relying on one person's unchecked impression.
Plan time for practice, independent coding, disagreement review, and clear documentation. State the reviewers' relevant training without implying certification they do not hold. If a project uses a simplified scheme, describe it honestly as that scheme.
Automated AU detection and landmarks
As of November 1, 2026, OpenFace describes landmark detection, head pose, AU recognition, and eye gaze. Its README directs commercial users to a separate commercial license through CMU. Check the published terms before treating an open repository as unrestricted commercial software.
As of November 1, 2026, Py-Feat describes AU, emotion, and landmark detection from images and video. Its code is MIT-licensed, while pretrained weights have separate terms. The toolbox's versions describe different landmark approaches, so record the version used.
The 2013 300-W resource uses Multi-PIE sixty-eight-point and fifty-one-point inner landmark markup. It describes normalized point-to-point landmark error and restricts its data to research, not commercial training. Landmark localization and AU interpretation should remain separate evaluation questions.
Limits and contextual review
Barrett and colleagues' 2019 review describes substantial variation across cultures, situations, and people. A facial configuration can accompany more than one emotion or none. That limits claims that any AU combination proves an internal state.
Jack and colleagues reported differing Western and East Asian representations of basic expressions in 2012. A subsequent reply argued that commonalities outweighed differences. Present the issue as an active scientific debate rather than declaring either universal certainty or complete meaninglessness.
Mental Edge surfaces time-coded expression and movement signals from recorded video for human review. Its category scores may indicate patterns worth inspecting alongside source moments. They do not prove feelings, intent, truthfulness, health, or suitability, and they should not replace independent coding.