Phrenology is not used in any scientific, medical, or clinical setting today. It has been classified as pseudoscience for nearly two centuries, and a thorough 2018 study using modern neuroimaging tools found zero evidence that the shape of the skull reveals anything about a person’s mental abilities or personality. The researchers concluded that “a more accurate phrenological bust should be left blank” because no regions on the head correlated with any of the mental faculties they tested. But while classical phrenology is dead, critics argue that its core logic has resurfaced in surprising places, particularly in artificial intelligence tools that claim to read character traits from faces.
What Phrenology Actually Claimed
Phrenology was developed in the early 1800s by the German physician Franz Joseph Gall. His central idea was that the brain is made up of distinct “organs,” each responsible for a specific mental trait, and that the strength of each trait could be measured by feeling the bumps and contours of a person’s skull. He eventually settled on 27 faculties, starting with basic drives like reproductive instinct and love of offspring, and working up to what he considered uniquely human traits like language and reasoning.
The system had a veneer of anatomical sophistication. Gall was a skilled neuroanatomist who correctly intuited that different parts of the brain handle different functions. But the leap from that general idea to reading personality from skull shape was enormous, and practitioners had no way to verify their claims. As early as 1815, critics dismissed the whole enterprise as “thorough quackery from beginning to end.”
How It Was Debunked
By the 1840s, phrenology had lost credibility among scientists in both Europe and the United States. A major blow came from French anatomists and physiologists led by Marie-Jean-Pierre Flourens, who performed brain experiments on animals and directly challenged the claim that specific skull regions mapped to specific mental traits. Elite interest in phrenology collapsed, though popular fascination lingered for decades longer.
The final nail came much later, when 21st-century researchers decided to give phrenology the most rigorous test it had ever received. Using modern brain-scanning technology on a large sample, they tested whether local scalp curvature could predict any mental capacity at all. It could not. The skull’s shape simply does not track with what the brain underneath is doing. The researchers noted that their cranial scanning methods might have one legitimate use: helping treat craniosynostosis, a condition where an infant’s skull bones fuse too early. That is as close as skull measurement gets to real medicine.
The One Idea That Survived
Phrenology got one thing roughly right, and modern neuroscience inherited it: the idea that different brain regions handle different functions. This concept, called localization, is foundational to how we understand the brain today. Brain scans reveal that language, vision, movement, and memory each depend on specific areas. The difference is that neuroscience measures brain activity directly, through blood flow, electrical signals, and metabolic changes, rather than inferring it from skull bumps.
Some critics have accused modern neuroimaging of being “a modern day version of phrenology” because brain scan studies sometimes just highlight which region “lights up” during a task without explaining what that activity means. But this criticism targets sloppy use of the tools, not the tools themselves. Well-designed imaging studies go beyond asking where activity occurs and investigate what functional role that activity plays. The methodology is fundamentally different from feeling someone’s head and declaring them honest or criminal.
Digital Phrenology and AI
The most troubling echo of phrenology today comes from machine learning. Over the past decade, a wave of studies and commercial products have claimed that algorithms can predict a person’s personality, mental health, sexual orientation, or even criminal tendencies by analyzing their face. Researchers and ethicists have given this trend a blunt label: “digital phrenology.”
The examples are striking. One widely discussed study claimed a neural network could predict sexual orientation from facial photos. Others have claimed to detect drug addiction, autism, and criminality from facial images alone. A separate project tried to identify deception during remote job interviews through facial analysis. Yet another study purported to predict political leanings from photographs. These efforts share phrenology’s foundational assumption: that external physical features reveal internal mental traits.
The criticism of these studies mirrors the original critique of phrenology. The algorithms often pick up on superficial cues like grooming, lighting, facial expression, or the way a photo was taken, rather than detecting any real biological signal. When researchers control for those confounding factors, the predictive power tends to collapse. A 2024 review published in the journal Patterns described this body of work as “AI pseudoscience,” “physiognomic AI,” and “junk science,” noting that the methods have been repackaged and deployed commercially despite lacking scientific validity.
Facial Analysis in Hiring
Where this becomes more than an academic debate is in the workplace. Companies have adopted AI-powered video assessment tools that analyze job candidates’ faces during interviews. One of the best-known, HireVue, used facial recognition not to verify identity but to evaluate what it called a candidate’s “cognitive ability,” “psychological traits,” “emotional intelligence,” and “social aptitudes” based on facial features and expressions during a recorded video interview. (HireVue later dropped its facial analysis component after sustained criticism.)
The concern is not just accuracy. These systems tend to absorb the biases present in their training data. If past hiring decisions reflected discrimination by race, gender, or age, the algorithm learns to replicate those patterns. As one analysis from Seattle University put it, the AI “mirrors our own stereotypes and biases, enforcing the occupational segregation loop.” The result is that both the hiring manager’s bias and the algorithm’s bias compound each other. Digital activists and legal scholars have pushed back against these tools, arguing that using biometric data to predict competency or personality is functionally the same error phrenologists made two centuries ago, just faster and at scale.
Why the Comparison Matters
Phrenology’s history is useful precisely because it shows how easily a scientific-sounding framework can be built on a flawed premise. Gall was not a con artist. He was a respected anatomist who made genuine contributions to understanding brain structure. But his system for reading character from skulls was untestable by the standards of his time, and when it finally was tested rigorously, it failed completely.
The same dynamic plays out with modern facial analysis AI. The tools are technically sophisticated, the datasets are large, and the results are presented with statistical confidence. But if the underlying premise is wrong, if faces do not actually encode personality or criminal intent, then no amount of computational power fixes the problem. It just makes the wrong answer more convincing. The academic community’s two-century rejection of phrenology is not just a historical curiosity. It is a warning about what happens when the desire to classify people outruns the evidence for doing so.