Human facial expressions represent one of the most intricate biological signaling systems in the natural world, operating as an immediate conduit for emotion, intention, and physiological state. To systematically deconstruct, quantify, and analyze these fleeting muscular movements without subjective bias, behavioral scientists rely on the standardized concept of the action unit. By parsing the face into anatomically distinct mechanical events, researchers have unlocked unprecedented empirical rigor in affective neuroscience, computational vision, and cross-cultural psychology.
Action Unit
1. Concise Definition
An action unit (commonly abbreviated as AU) is an anatomically grounded, standardized unit of visible facial muscle movement designated within the Facial Action Coding System (FACS). Rather than describing emotional constructs such as happiness or sorrow, an action unit delineates the discrete physical contraction or relaxation of a specific muscle or muscle group that produces an observable alteration in facial morphology.
In behavioral science, each action unit is cataloged numerically (e.g., AU1 for Inner Brow Raiser, AU12 for Lip Corner Puller) to afford an objective, descriptive taxonomy. This taxonomic framework ensures that researchers record visible dynamic cues—such as furrowing, wrinkling, deepening of sulci, or shifting of facial landmarks—independently of any higher-order psychological interpretation. By decoupling physical movement from affective inference, the construct provides a foundational common language for both manual behavioral coding and automated computer vision algorithms.
2. Etymology & Linguistic Origin
The term action unit derives from the intersection of classical kinetic terminology and operational scientific classification. The noun action traces back through Old French to the Latin actio (a doing, performing, or lawsuit), rooted in agere (to set in motion, drive, or do). The term unit originated from the Latin unitas (unity, oneness), drawn from unus (one), reflecting an indivisible elemental standard of measurement.
The compound nomenclature was formally coined in the late 1970s by American psychologists Paul Ekman and Wallace V. Friesen during the development of FACS. Recognizing that single facial muscles do not always act in total mechanical isolation—and conversely, that complex muscles such as the orbicularis oris or frontalis can act differentially in distinct regions—Ekman and Friesen opted against the term “muscle unit.” They selected action unit to emphasize an observable, functional biomechanical performance rather than purely invisible neuromuscular innervation.
3. Pronunciation & Grammatical Form
In standard English phonetics, action unit is pronounced /ˈæk.ʃən ˈjuː.nɪt/. It functions grammatically as a compound countable noun, taking the regular plural form action units. Within technical academic manuscripts, it frequently appears as an attributive noun adjunct, as seen in phrases such as “action unit detection,” “AU co-occurrence matrices,” and “AU-level intensity ratings.”
4. Detailed Conceptual Explanation
At its core, the action unit operationalizes the human face as a biomechanical display canvas governed by approximately forty individual muscles. These muscles are biologically distinctive because they insert directly into the superficial fascia and dermis rather than into opposing skeletal structures. Consequently, their contractions pull the skin toward fixed cranial anchoring points, generating transient folds, bulges, flattening, and structural displacements. An action unit captures each of these discernible changes systematically.
The scope of action units extends beyond isolated, unilateral contractions to encompass coordinated kinematic patterns. For instance, AU1 corresponds to the contraction of the medial belly of the frontalis muscle, elevating the inner corners of the eyebrows and generating horizontal furrows restricted to the center of the forehead. Conversely, AU2 recruits the lateral belly of the same muscle, pulling the outer eyebrows upward. By classifying these distinct functional vectors of a singular anatomical muscle into separate action units, the taxonomy achieves unmatched descriptive resolution.
Furthermore, action units are explicitly non-interpretive. Coding an AU4 (Brow Lowerer) indicates strictly that the corrugator supercilii, depressor supercilii, and procerus have converged to draw the brows downward and together, forming vertical wrinkles between them. It does not dictate that the subject feels anger, concentration, confusion, or physical pain. This structural boundary is crucial: the action unit operates at the somatic level of description, serving as the empirical substrate from which cognitive, affective, and communicative hypotheses can subsequently be tested.
Beyond spatial coordinates, action units incorporate temporal and intensity parameters. The standard protocol annotates intensity along a five-point ordinal continuum from trace (A) to maximum (E), capturing minute variations ranging from subconscious micro-expressions to pronounced theatrical displays. Temporally, an action unit follows a tripartite trajectory consisting of onset (the beginning of movement), apex (the period of peak intensity), and offset (the return to neutral morphology).
5. Historical Development
The scientific lineage of describing facial behavior systematically begins with French anatomist Duchenne de Boulogne in the mid-19th century. Duchenne utilized localized faradic electro-stimulation on living subjects to identify which muscles generated specific emotional grimaces, producing photographic plates published in his 1862 work The Mechanism of Human Facial Expression. Charles Darwin immediately adopted Duchenne’s work in his seminal 1872 treatise The Expression of the Emotions in Man and Animals, arguing that facial actions were evolved, adaptive vestiges common across human populations.
Despite these early historical foundations, the mid-20th century lacked an objective methodology to record these dynamic movements reliably. Anthropologists and psychologists frequently relied on arbitrary descriptive labels such as “smirk,” “grimace,” or “frown,” introducing subjective interpreter bias. In the late 1960s, Swedish anatomist Carl-Herman Hjortsjö attempted to bridge this gap by publishing Man’s Face and Mimic Language (1969), illustrating how distinct muscle actions shape visual appearance.
Building directly upon Hjortsjö’s preliminary structural descriptions, Paul Ekman and Wallace V. Friesen spent nearly a decade rigorously mapping self-induced facial muscle contractions to visual consequences. In 1978, they formally published the Facial Action Coding System, establishing the action unit as the definitive baseline metric for empirical research. The system was comprehensively updated in 2002 in collaboration with Joseph C. Hager, formalizing digital measurement criteria, head and eye orientation units, and fine-grained intensity scoring rules.
6. Theoretical Foundations
The concept of the action unit is theoretically situated at the nexus of evolutionary biology, biomechanics, and affective neuroscience. Primarily, it underpins Ekman’s Basic Emotion Theory (BET), which posits that a set of discrete, phylogenetically conserved internal emotional states (such as fear, anger, disgust, sadness, surprise, and joy) trigger innate, universal motor programs executed via coordinated configurations of action units.
Conversely, the action unit serves as an indispensable tool for opposing paradigms, such as the Constructed Emotion Theory championed by Lisa Feldman Barrett. Within Barrett’s framework, facial movements do not represent uniform, hard-wired expressions of basic emotional categories; rather, they are variable physical manifestations synthesized by the brain to match contextual predictions. Because action units offer a neutral linguistic framework detached from emotional labels, constructivist researchers utilize them to document intra-category variability and challenge prototypical expression archetypes.
Furthermore, action unit mechanics align with the Facial Feedback Hypothesis, which suggests that skeletal muscle feedback from facial actions modulates subjective emotional experience. By isolating specific action units—such as differentiating the zygomaticus major activation of AU12 from the simultaneous orbicularis oculi contraction of AU6—theorists can rigorously examine whether the physical morphology of an expression actively shifts autonomic nervous system arousal and neural activation patterns.
7. Key Components, Types & Dimensions
Action units are structured across discrete anatomical categories, encompassing movements throughout the upper and lower face, as well as auxiliary communicative behaviors:
- Upper Face Action Units: Primarily involve the ocular, orbital, and forehead regions. Notable examples include AU1 (Inner Brow Raiser), AU2 (Outer Brow Raiser), AU4 (Brow Lowerer), AU5 (Upper Lid Raiser), AU6 (Cheek Raiser), and AU7 (Lid Tightener).
- Lower Face Action Units: Encompass the mouth, lips, chin, and lower cheeks. They are subdivided into horizontal actions (e.g., AU14 Dimpler, AU20 Lip Stretcher), oblique actions (e.g., AU12 Lip Corner Puller), vertical actions (e.g., AU10 Upper Lip Raiser, AU15 Lip Corner Depressor), and orbital/protruding actions (e.g., AU18 Lip Puckerer, AU24 Lip Pressor).
- Action Descriptors (ADs): Facial movements that lack localized, isolated muscular specificity or involve non-muscle tissue behavior, such as AD19 (Tongue Show), AD29 (Jaw Thrust), or AD32 (Bite).
- Head and Eye Orientation Units: Additional codes capturing gaze trajectory and head tilt (e.g., AU51 Head Turn Left, AU64 Eyes Downward), which contextualize the social and emotional meaning of concurrent facial actions.
- Intensity Dimension (A–E Scale): A five-tiered metric evaluating the degree of physical involvement: A (Trace), B (Slight), C (Marked/Pronounced), D (Severe/Extreme), and E (Maximum) mechanical displacement.
- Unilateral vs. Bilateral Dimension: Coding designations identifying whether an action unit manifests symmetrically across both facial hemispheres or presents unilaterally with distinct left/right asymmetries.
8. Examples & Illustrative Cases
A quintessential illustration of action unit configuration is the physiological differentiation of smiling, initially distinguished by Duchenne de Boulogne and formalized via FACS. A spontaneous, emotionally congruent expression of enjoyment—often termed the Duchenne smile—is defined by the synchronous bilateral activation of AU12 (Lip Corner Puller, driven by the zygomaticus major) and AU6 (Cheek Raiser, driven by the orbital portion of the orbicularis oculi). When an individual displays a polite, socially managed, or non-felt smile, AU12 frequently occurs in total isolation, leaving the periorbital tissue relaxed and omitting the characteristic crow’s-feet wrinkles and infraorbital cheek lifting induced by AU6.
A second clinical case involves the diagnosis of clinical depression and somatic pain. Patients enduring severe physical distress routinely display an involuntary affective grimace consisting of the composite bundle: AU4 (Brow Lowerer), AU6/AU7 (Eye Narrowing), AU9/AU10 (Nose Wrinkling and Upper Lip Elevation), and AU43 (Eyes Closed). By quantifying the duration and intensity of this action unit constellation, medical practitioners can objectively gauge acute discomfort in non-verbal infants or post-stroke patients who are otherwise unable to communicate on subjective numerical rating scales.
9. Measurement & Assessment
The historical gold standard for measuring action units is manual visual inspection performed by a Certified FACS Coder. Achieving certification requires passing the rigorous Facial Action Coding System Final Test administered through accredited research laboratories. Coders evaluate high-definition video footage frame-by-frame, often slowed to 30 or 60 frames per second, documenting every onset, apex, offset, and intensity level. Although exceptionally accurate, manual coding is profoundly labor-intensive, typically demanding up to two hours of human analysis for every sixty seconds of recorded behavioral interaction.
To overcome this temporal bottleneck, the field has increasingly transitioned toward automated measurement frameworks leveraging computer vision and deep neural networks. Modern platforms such as OpenFace, Affectiva, and specialized convolutional neural network (CNN) architectures utilize landmark alignment to calculate geometric changes and texture deformations across the face. These automated systems predict continuous action unit probabilities across extensive longitudinal video datasets, dramatically accelerating research output.
A complementary physical measurement modality is Facial Electromyography (fEMG). Surface electrodes placed strategically over muscle bellies (e.g., the corrugator supercilii for AU4 or the zygomaticus major for AU12) measure localized neuromuscular electrical activity. Unlike visual coding, fEMG can register micro-contractions that fail to alter visible skin geometry, providing valuable data on subliminal psychological reactions that do not meet the minimum physical threshold for visual action unit recognition.
10. Applications & Practical Significance
The operational precision of action units has yielded substantial practical applications across clinical psychiatry, human-computer interaction, neuromarketing, and legal forensics. In psychiatric diagnostics, researchers evaluate blunted or incongruent affect in conditions like schizophrenia and major depressive disorder by measuring the reduced frequency, delayed latency, and low intensity of expressive action units during clinical interviews.
In technological engineering, action unit recognition powers intelligent human-computer interfaces. Advanced driver-monitoring systems monitor occurrences of AU41/42 (Eye Droop/Slit) and AU43 (Eyes Closed) alongside AD45 (Blink Rate) to detect dangerous operator fatigue and initiate real-time safety interventions. Similarly, affective computing platforms adapt digital pedagogical content in real time when users exhibit persistent AU4 (Brow Lowerer), signaling cognitive overload, confusion, or visual strain.
Within consumer research, continuous action unit modeling replaces subjective self-report questionnaires to track real-time audience engagement. Measuring transient displays of AU14 (Dimpler) or AU9 (Nose Wrinkler) while participants view commercial media provides clear, second-by-second indicators of skepticism, aversion, or authentic amusement.
11. Research & Empirical Evidence
Decades of empirical studies validate the ecological and neurological validity of action units. Foundational cross-cultural fieldwork conducted by Paul Ekman and Wallace V. Friesen in Papua New Guinea demonstrated that pre-literate, isolated Fore tribesmen accurately linked specific action unit configurations—such as AU1+AU4+AU15—with universal narrative scenarios describing sadness and loss. These findings provided historic empirical weight to the argument for biological universality in facial kinetics.
In evolutionary psychology, Frans de Waal and Bridget Waller extended this paradigm to primates by constructing the ChimpFACS system. Waller and her colleagues mapped chimpanzee facial musculature directly to human action units, illustrating that non-human primates share homologous action units (such as AU12 equivalents during play sessions), proving deep phylogenetic continuity in mammalian social signaling.
Recent advances in computational modeling have evaluated millions of naturalistic facial interactions worldwide. In a milestone study published in Nature, Alan Cowen and colleagues analyzed over six million video clips from 144 nations using automated action unit models. Their data revealed that sixteen distinct action unit profiles systematically co-occur with contextual emotional events across divergent geographic cultures, confirming robust cross-cultural universality within real-world environments.
12. Cultural & Cross-Cultural Considerations
Although the anatomical capacity to execute specific action units is universal among healthy human populations, their social deployment, baseline frequency, and contextual display rules are heavily mediated by culture. Seminal work by David Matsumoto demonstrated that cultural display rules dictate whether individuals amplify, de-amplify, mask, or entirely suppress particular action units in public social settings.
Research led by Rachael Jack at the University of Glasgow has highlighted cross-cultural differences in how facial movements are visually processed. While Western Caucasian observers typically distribute their visual attention across both the upper (AU1, AU2, AU4) and lower facial regions (AU12, AU20) to interpret social meaning, East Asian participants tend to focus fixations on the orbital ocular region (AU6, AU7). Consequently, observers from distinct linguistic and cultural traditions often weight specific action units differently when decoding ambiguous human intentions.
13. Criticisms, Debates & Limitations
Despite its widespread utility, the action unit construct faces significant theoretical and practical criticisms. A prominent debate involves the biological assumption that human emotion can be reduced to static combinations of action units. Critics argue that real-world social interaction involves dynamic, contextual bodily communication that cannot be fully captured by isolating isolated facial muscle twitches on a standardized grid.
A second critique addresses the physical constraints of visual coding. Facial morphology varies widely across age, sex, adipose distribution, and ethnicity. Aging skin displays baseline structural wrinkles that can mimic traces of AU4 or AU6 even at absolute muscular rest, leading to potential coding inaccuracies. Furthermore, facial hair, heavy makeup, and non-standard camera angles can obscure visual landmarks, posing ongoing engineering challenges for automated action unit recognition models.
Finally, researchers highlight the operational ambiguity surrounding action unit co-occurrence. When multiple muscles contract simultaneously, their mechanical forces interact non-linearly, frequently masking or distorting individual movement cues. This physical masking makes it difficult for both human coders and computer algorithms to verify whether a constituent action unit is genuinely engaged, underscoring the ongoing need for rigorous multi-sensor validation.
14. Related Terms & Distinctions
- Micro-expression: A brief, involuntary facial expression lasting only a fraction of a second (typically 1/25 to 1/5 of a second) revealing genuine suppressed emotion, composed of rapid, low-duration action units.
- Facial Action Coding System (FACS): The overarching comprehensive manual and taxonomic methodology within which action units represent the fundamental descriptive measurement elements.
- Macro-expression: A typical visible facial movement lasting between 0.5 to 4 seconds, composed of action units that match normal social interaction and conversation.
- Action Descriptor (AD): A FACS classification category dedicated to gross facial, jaw, or tongue movements that cannot be attributed to a specific single facial muscle group.
- Facial Landmark: Geometric coordinates (such as the corners of the mouth or pupils) tracked in computer vision algorithms to detect and measure the motion of underlying action units.
15. Summary & Key Takeaways
The action unit is the definitive empirical standard for quantifying facial kinematics, transforming the subjective interpretation of facial displays into an objective, repeatable science. By delineating human facial movement through its underlying anatomical mechanisms, the action unit bridges fundamental biomechanics with higher-order psychological, clinical, and computational inquiries.
Whether annotated by specialized human analysts or parsed by deep-learning models, action units remain essential for deciphering the structural mechanics of human emotion, intentionality, and social interaction worldwide.
References
- Cowen, A. S., Keltner, D., Schroff, F., Jou, B., Adam, H., & Prasad, G. (2021). Sixteen facial expressions occur in similar contexts worldwide. Nature, 589(7841), 251–257. https://doi.org/10.1038/s41586-020-3037-7
- Darwin, C. (1872). The expression of the emotions in man and animals. John Murray.
- Duchenne de Boulogne, G. B. (1862). The mechanism of human facial expression (R. A. Cuthbertson, Trans., 1990). Cambridge University Press.
- Ekman, P., & Friesen, W. V. (1978). Facial Action Coding System: A technique for the measurement of facial movement. Consulting Psychologists Press.
- Jack, R. E., Blais, C., Scheepers, C., Schyns, P. G., & Caldara, R. (2009). Cultural confusions show that facial expressions are not universal. Current Biology, 19(18), 1543–1548. https://doi.org/10.1016/j.cub.2009.07.051
- Waller, B. M., Vick, S. J., Parr, L. A., Bard, K. A., Pasqualini, M. C., Gothard, K. M., & Dunbar, R. I. (2006). Intramuscular electrical stimulation of facial muscles in humans and chimpanzees: Duchenne revisited and extended. Emotion, 6(3), 367–382. https://doi.org/10.1037/1528-3542.6.3.367