Overview
This page describes face reading as a family of practices rather than one method. It sets out where the Chinese, Greek and European, Indian, Arabic and Japanese traditions came from, how they differ, what psychology and computer science research has and has not found, and what the historical record says about harm, including discrimination and the misuse of facial analysis software.
What types of Face Reading are there?
| Type | How it differs |
|---|---|
| Chinese Mian Xiang | Chinese tradition with surviving manuals from the Dunhuang manuscripts and the Song period, which analyzes body and face by the five phases, animal resemblance, complexion and life periods. |
| Western physiognomy | European line running from Greek treatises attributed to Aristotle through della Porta in 1586 and Lavater, which assigns traits by resemblance to animals, ethnicity and expression. |
| Indian Samudrika Shastra | Sanskrit body-reading texts, such as the Brihat Samhita chapter on the features of men, that link limbs, palm lines, face, voice and gait to fortune, status and lifespan. |
| Ancient Greek physiognomonics | Greek treatises, including the Physiognomica attributed to Aristotle's followers, that judged character from the body by resemblance to animals, ethnicity and facial expression. |
| Lavater's physiognomy | Eighteenth-century systematization by the Swiss theologian Johann Caspar Lavater, who promoted physiognomy through essays, lectures and public debate. |
| Criminal anthropology | Nineteenth-century approach of Cesare Lombroso, who held that criminal traits were innate and visible in features such as the brow, jaw and upper lip, a view later discredited. |
| Arabic firasa | Arabic-language term with four meanings, one of which became physiognomy proper after Greek treatises, including one attributed to Polemon, were translated and circulated around the ninth century. |
| Japanese physiognomy | Japanese practice built largely on the Chinese Shenxiang quanbian, which was first published in Japan in 1651 and is still used in abridged form as a standard manual. |
| Traditional Chinese medicine facial inspection | Diagnostic method within Chinese medicine that examines facial color and gloss as part of inspection, kept distinct from fortune-reading physiognomy despite shared vocabulary. |
| Facial action coding and expression research | Experimental psychology that codes facial muscle movements with the Facial Action Coding System and tests what they reveal about emotion, without claims about fate or character. |
| First-impression research on facial trait inference | Experimental research by Todorov and colleagues on how quickly people infer traits such as competence from faces and how those judgments shape choices, with accuracy questioned. |
| AI facial analysis | Machine learning systems trained on face photographs to classify traits such as sexual orientation or criminality, whose results critics attribute to dataset artifacts and presentation cues. |
Where does Face Reading come from?
Separate lineages with a shared premise
Face reading is a family of practices that infer character, fortune or health from the face and body. The sources reviewed describe lineages that developed in different regions and languages, among them Chinese, Greek and later European, Indian, Arabic-language and Japanese traditions, and they do not show a single common origin. The National Library of Medicine exhibition on physiognomy traces the European line to ancient Greece, where philosophers and physicians treated the body as a guide to the soul and assigned traits by resemblance to animals, ethnicity and facial expression.
Chinese lineage
Livia Kohn's 1988 study in Asian Folklore Studies cites a passage of the Zuozhuan as one of the earliest known cases of a professional physiognomist in China. She reports that the first systematic manual surviving today is the Xiangshu among the Dunhuang manuscripts, ascribed to Xu Fu of the Han dynasty. From the tenth century onward, she finds, physiognomic materials increase and the systems grow more complex, and she traces a unified tradition from the early Song period to her own day.
Kohn reports that the later standard textbook, the Shenxiang quanbian, was compiled by the court physiognomist Yuan Zhongche in the early Ming period. The tradition names the tenth-century Daoist Chen Tuan as its founder, but she judges the evidence from the physiognomic tradition itself to be spurious. A Wellcome Collection record describes a 1807 Japanese woodblock edition and dates the first Japanese edition to 1651. A conference paper dates the Divine Fortuneteller Ma Yi to the Northern Song period, 960 to 1127 CE.
Western lineage, criminology and race science
According to the NLM exhibition, Giovanni Battista della Porta helped revive physiognomy in 1586 with De humana physiognomonia, and the Swiss theologian Johann Caspar Lavater systematized it in his Essays on Physiognomy. The exhibition describes Cesare Lombroso, whose L'Homme criminel appeared in 1887, as holding that criminal traits were innate and visible in the face, and says these ideas held currency until World War II.
The exhibition places physiognomy within eighteenth- to twentieth-century physical anthropology that used face, skull and body measurements to define race and rank groups. It says Francis Galton brought physiognomy into eugenics with composite portraits in 1878, and that by the mid-twentieth century physiognomy was discredited.
Indian lineage
The Indian material is documented here mainly through the Brihat Samhita attributed to Varahamihira. The online edition of N. Chidambaram Iyer's 1884 English translation dates it to the sixth century CE and also gives a first-century BCE alternative. Its chapter 68, on the features of men, applies body-reading to fortune and status. The sources opened do not give a scholarly history of the Samudrika Shastra as a named body of texts, so its dating and links to the Brihat Samhita remain open.
What do the names and terms mean?
Names and scope
Physiognomy is derived, in Mohammed Ghaly's account, from Greek words for nature, law and judge or interpreter, and it denotes judging character from bodily features. Face reading is the broader popular label. Mian Xiang and Mien Shiang are treated here as alternative romanizations of Chinese face reading, a usage the open sources reviewed do not themselves define. Kohn's study uses the broader Chinese term xiang, as in Xiangshu and Shenxiang quanbian, and describes an analysis of the whole body that also covers bones, voice and complexion.
Arabic firasa
Ghaly reports four meanings of the Arabic term firasa: sharp-mindedness, tracking and kinship judgment (qiyafa), a mystical intuition said to be given by God, and a sense equivalent to Greek physiognomy. Only the last concerns reading bodily features, so the Arabic term covers more than face reading.
He writes that the physiognomic sense depended on translations of Greek treatises, including two books attributed to Aristotle and one to Polemon, which began to circulate around the ninth century CE. He adds that Indian and Persian influences are possible but seem to have faded once the Greek sources were translated.
Related terms kept distinct
Kohn notes that complexion (se) is both a physiognomic term and a technical term in the classics of traditional Chinese medicine. A 2014 study describes facial complexion inspection as an intuitive diagnostic method within that medical system. This page keeps medical facial inspection separate from fortune-reading physiognomy even though the vocabulary overlaps.
The Facial Action Coding System, used in expression research, is described by Barrett and colleagues as purely descriptive. It records which facial movements occur and is agnostic about whether they express emotions or any other mental event.
What does Face Reading involve?
Chinese methods
Kohn describes the Mirror of Auras, ascribed to Chen Tuan, as proposing a theory of interaction between body and spirit. It gives rules for analyzing body types and features by the five phases and by animal types, and it analyzes facial complexion by the five phases. She adds that parts of the face are identified with periods of life, and that the practitioner also judges the overall aura, the energy (qi) and the color.
The authors of a paper on the Divine Fortuneteller Ma Yi report that it lists 35 body parts and sorts bodies into five types named for metal, wood, water, fire and soil. They say it compares three vertical facial distances by eye and ties the ears to ages 1 to 14, the nose to 40 to 50 and the jaw to after 60. Their framing of this as early anthropometry is their own interpretation. The sources reviewed do not describe the twelve palaces or left and right halves.
Indian and European methods
Iyer's translation of Brihat Samhita chapter 68 names twelve characteristics to examine, including complexion, voice, strength, height, weight, disposition and gait. It links soft, well-formed feet to kingship, reads palm and forehead lines, and says a face resembling a lion or cow suggests strength while one resembling an ass or camel suggests misfortune.
In the European line, the NLM exhibition says Greek texts assigned traits by animal resemblance, ethnicity and expression, and that Lombroso pointed to a sloping brow, large jaw and thin upper lip as signs of an atavistic criminal type.
Coded and computational methods
Barrett and colleagues describe human coders who train for many weeks to apply the Facial Action Coding System, as well as automated detection of facial actions from photographs and video. A 2018 study extracted features from 35,326 facial images with deep neural networks and entered them into a logistic regression to classify sexual orientation.
What is Face Reading used for?
Traditional claims
Kohn writes that in Chinese history heirs apparent were nominated on the advice of physiognomists, and that officials or nobles were demoted or killed on the basis of such predictions. Iyer's translation of the Brihat Samhita ties features to wealth, sons, kingship, poverty and lifespan. Practitioners of Chinese medicine have used facial color as part of diagnosis, which the 2014 study describes as depending on the practitioner's experience.
Modern claims
Barrett and colleagues survey the assumption that emotional state can be read from facial movements and report that it influences legal judgments, security protocols, education, commercial applications and psychiatric care. They also report that agents of two US federal agencies were trained to detect emotions and intentions from facial configurations. Recent machine-learning papers have claimed that sexual orientation or criminal status can be classified from face photographs.
What does the research show?
First impressions from faces
A 2005 Science study found that competence judgments made from facial appearance alone predicted 68.8 percent of 2004 US Senate races and tracked margins of victory, with judgments forming within a one-second glimpse. This shows that people form and act on quick trait judgments, not that the judgments are accurate. A 2015 Annual Review of Psychology article reports broad agreement among observers about traits seen in faces, and argues that the diagnostic validity of those attributions has been substantially overstated.
Facial expression and emotion
A 2019 review in Psychological Science in the Public Interest examined evidence on anger, disgust, fear, happiness, sadness and surprise. It found that people sometimes smile when happy and frown when sad more than chance would predict, yet expression varies substantially across cultures, situations and individuals, and one configuration of movements can express more than one emotion category. The authors conclude that facial movements convey a range of social information but do not reliably reveal a specific emotional state on their own.
Traditional claims and health inspection
The sources reviewed include no controlled test of the Chinese, Indian or Arabic claims linking facial features to fortune, character or lifespan. The Ma Yi paper compares its facial landmarks with modern anthropometry, which describes measurement, not validity. In a 2014 computer-vision study of 122 participants, a color model matched physician-assigned facial color labels at 86.89 percent. The authors call the work preliminary, and it does not test whether facial color identifies disease.
Machine learning on faces
The 2018 study reported classifying sexual orientation from a single image in 81 percent of cases for men and 71 percent for women, with features that included transient ones such as grooming style. A 2019 replication on 20,910 photographs reported lower accuracy for men and said it remained unclear how much came from facial structure versus grooming, presentation and lifestyle. A 2020 critique argues that reported accuracy of up to 97 percent for criminality from faces reflects confounds in the datasets, such as criminal and non-criminal images differing in file format and source.
What are the safety considerations?
Discrimination and pseudoscience
The NLM exhibition documents physiognomy's use in scientific racism and eugenics, including Galton's composite portraits and Lombroso's claims about the criminal face, and notes that the Nazi regime's use of eugenics led to its widespread denunciation. The authors of the 2020 critique warn that belief in an algorithm for criminality from faces carries a large social cost, and they state that nothing about committing a crime causes a distinctive change in facial appearance.
Privacy, consequential decisions and health
The 2018 sexual orientation study and its 2019 replication both warn that facial classification threatens the privacy and safety of gay men and women. Because the evidence reviewed shows limited validity for trait and emotion inferences from faces, these sources do not support using them as a basis for decisions about a person. A study of juristic discussions in Islamic law, published in 2009, examines how physiognomy intersected with the image of people with disabilities.
The sources reviewed contain no evidence that reading facial signs detects disease. Facial inspection within Chinese medicine is described as subjective and experience-based. It does not replace medical assessment, and symptoms or health concerns call for a licensed clinician. No statutory regulation or training standard for face reading, and no professional register, was found in the sources reviewed, which were searched for Chinese, Indian, Arabic, Western and expression-research material.
Connected Entries
Classification and context
- Entry type
- Modality
- Practice type
- Family of face-based character, fortune and health interpretation traditions
- Traditions
- Chinese physiognomy (xiangshu), Greek and European physiognomy, Indian body-reading texts (Samudrika Shastra), Arabic firasa, Japanese physiognomy, Traditional Chinese medicine facial inspection
- Concepts
- Five phases body types, Animal resemblance, Facial complexion, Body-spirit interaction, Facial trait inference, Facial action coding, Criminal anthropology
- Categories
- Metaphysical, Alternative
Sources and discovery context
Research and public-health sources support evidence and safety context. Directory and discovery sources describe navigation, relationships, practitioners, and access—not clinical efficacy.
A free review concluding that facial movements alone do not reliably reveal specific emotional states across contexts, people and cultures.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports statements about facial expression and emotion in six emotion categories and the descriptive nature of FACS. It does not address personality, fortune or health readings of faces and must not be extended to them.
- Supports sections
- Terminology, Practice Description, Claimed Uses, Research and Evidence
Reports that judgments from facial appearance alone predicted 68.8 percent of 2004 Senate races within a one-second exposure.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the statement that rapid competence judgments from faces predicted US congressional election outcomes above chance. It does not show that the judgments reflect true competence.
- Supports sections
- Research and Evidence
A review of how people attribute traits to faces that argues the accuracy of such attributions has been greatly overstated.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the statements about rapid agreement among observers and the authors' argument that diagnostic validity is overstated. It does not assess any named face reading tradition.
- Supports sections
- Research and Evidence, Safety and Suitability
Reports classifier accuracy of 81 percent for men and 71 percent for women from one image and warns of risks to privacy and safety.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the sample size, reported accuracy, inclusion of transient features such as grooming, and the authors' privacy warning. It must not be read as showing that facial structure alone determines orientation.
- Supports sections
- Practice Description, Claimed Uses, Research and Evidence, Safety and Suitability
A preprint replication that reports lower accuracy than the original and leaves open how much comes from facial structure versus grooming and presentation.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the 20,910-photograph sample, the lower accuracy for men, the statement that sources of the signal remain unclear, and the safety warning. It does not settle what the classifiers detect.
- Supports sections
- Research and Evidence, Safety and Suitability
Argues that criminality-from-face algorithms rest on a flawed concept and that high reported accuracy comes from confounded datasets.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the claim that reported accuracy in the criticized studies is attributable to dataset confounds and the warning about social cost. It does not evaluate the other AI classification studies cited on the page.
- Supports sections
- Claimed Uses, Research and Evidence, Safety and Suitability
A government history exhibition tracing physiognomy from Greek texts to eugenics and modern artificial intelligence.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports dates and descriptions for Greek, Renaissance, Lavater, Lombroso, Galton and scientific racism material. It is a curated exhibition and does not cover Chinese, Indian or Arabic traditions.
- Supports sections
- History and Origins, Practice Description, Safety and Suitability
A scholarly study of the Chinese physiognomy tradition from the Dunhuang manuscripts through the Song and Ming periods.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports statements about Chinese physiognomy manuals, the five phases and animal types, and the uncertain attribution to Chen Tuan. It describes tradition, not validity, and says nothing about the 'three zones' or 'twelve palaces' taxonomies.
- Supports sections
- History and Origins, Terminology, Practice Description, Claimed Uses
A conference paper describing the measurement landmarks and body types in a Northern Song physiognomy manual.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the book's date, its five body types, its 35 body parts and its age-region associations. The authors' view that it has anthropometric value is their interpretation and must not be read as evidence of predictive validity.
- Supports sections
- History and Origins, Practice Description, Research and Evidence
A library record for a 1807 Japanese woodblock edition of the Shenxiang quanbian, licensed CC BY 4.0.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the 1807 Japanese edition, the 1651 first Japanese edition and the five-element and zodiac-animal classification. It does not describe Japanese practice in general.
- Supports sections
- History and Origins
An article that distinguishes four meanings of firasa and traces the physiognomic sense to translated Greek treatises.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the four meanings of firasa and the role of translated Greek texts. It does not describe how firasa was practiced in detail, and its legal analysis was not summarized.
- Supports sections
- Terminology, Safety and Suitability
A nineteenth-century English translation of a Sanskrit chapter that links bodily features to fortune and status.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports what this chapter says about judging fortune from bodily features. The chapter does not use the term Samudrika Shastra in the portion read, and it must not be read as describing that tradition as a whole or its later manuals.
- Supports sections
- History and Origins, Practice Description, Claimed Uses
The index of an online translation that lists the work's author and a stated date range.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports only the attribution to Varahamihira and the stated date range. It does not establish when body-reading traditions began in India.
- Supports sections
- History and Origins
A study of 122 participants that models facial color classes against physician labels and describes the subjectivity of traditional facial inspection.
How this source relates
- Source link
- Exact source page
- Relationship
- Supports the description of facial complexion inspection as subjective and experience-based, and the 122-participant model result. It does not show that facial color identifies disease and says nothing about physiognomy for fortune or character.
- Supports sections
- Terminology, Claimed Uses, Research and Evidence, Safety and Suitability