Pareidolia: Why Your Brain Sees Faces in Everything

People have been finding faces where there aren’t any for as long as we’ve been human. In 1978, a woman in New Mexico noticed a burn mark on a tortilla that looked exactly like the face of Jesus. She kept it. Thousands of people came to see it. In 2004, a grilled cheese sandwich with what appeared to be the Virgin Mary sold on eBay for $28,000. In the 1990s, people swore they could see a demonic grin hidden in Princess Diana’s hair in certain photographs. And in 2006, a Doritos crisp supposedly bearing the face of Pope John Paul II was listed for auction.

None of these were faked. The people who saw them genuinely saw them. Their brains were doing something completely normal.That something has a name: pareidolia.

What Is Pareidolia?

Pareidolia (pronounced pair-eye-DOH-lee-ah) is a psychological and neurological phenomenon in which the brain perceives a meaningful pattern, usually a face or figure, within a random or ambiguous stimulus. It happens with visual information most often, but it occurs with sound too.

The word comes from Greek. “Para” means beside or alongside, and “eidolon” means image or form. Together they point to a secondary image seen alongside the real one, a perception that rides on top of something else.

If you’ve looked at the sofa image at the top of this page and seen two eyes staring back at you, that’s pareidolia. Nothing unusual happened. Your brain just ran its standard operating procedure.

The specific term for face detection in ambiguous stimuli is prosoppareidolia, from the Greek “prosopon” meaning face. It’s the most studied and most common form, though pareidolia extends to other patterns too: figures in clouds, shapes in rock formations, words in static noise.

The Evolutionary Case: Why the Brain Does This

This is the more important question. Pareidolia isn’t a glitch. It’s a feature, and understanding why requires going back a very long way.

Over millions of years of evolution, the human brain faced a recurring problem: identifying threats quickly. Spotting a predator’s face in undergrowth, or reading the expression of a rival in low light, was directly tied to survival. Those who detected face-like patterns faster and more reliably had a better chance of staying alive. Their genes persisted.

The result is a visual system with an unusually high sensitivity to face-like configurations. That sensitivity is so finely tuned that the brain will happily identify a face that isn’t there rather than miss one that is.

In statistical terms, this is a Type I error, a false positive. The cost of seeing a face that doesn’t exist is minimal: a second of unnecessary alertness. The cost of failing to see a face that does exist could be death. The brain ran the maths and calibrated itself accordingly.

This logic was formalised by researchers including Arne Öhman and others working within Error Management Theory (EMT). EMT predicts that under uncertainty, cognitive systems will bias towards errors with lower costs. In human evolutionary history, the false positive was cheap. The false negative was potentially fatal. So the system errs heavily on the side of seeing faces.

This is also why pareidolia is stronger in conditions of anxiety, sleep deprivation, or high arousal. When threat-detection systems are running hot, the tendency to find meaningful patterns in ambiguous information increases.

The Neuroscience: What Actually Happens in the Brain

When you see a face, real or imagined, a specific region of your brain is doing most of the work.

The fusiform face area (FFA), located in the fusiform gyrus of the temporal lobe, was first identified by Nancy Kanwisher and colleagues at MIT in the 1990s. Their finding was striking: this region responds selectively to human faces, not to other objects. It activates when you see a face in a way it simply doesn’t for chairs, cars, or animals.

The key insight for pareidolia came from later research. The FFA doesn’t just respond to real faces. It responds to face-like configurations. A 2014 study published in the journal Cortex showed that when participants perceived a face in an ambiguous image, the same neural regions that activated for real faces also activated for the illusory ones, though with weaker signal. The pattern of activation was structurally identical. The brain wasn’t treating these as fundamentally different categories.

Research by neuroscientist Nouchine Hadjikhani at Harvard found that the visual system interprets stimuli rapidly and pre-consciously. Before you consciously decide what you’re looking at, part of your brain has already found the pattern. The FFA runs ahead of awareness.¹

Work by Carlos Mondragón and colleagues using fMRI also demonstrated frontal lobe involvement in pareidolia, not just visual processing regions. This matters. The frontal lobe is where prior knowledge, beliefs, and expectations live. Its involvement confirms that what you already think influences what your visual system extracts from incoming data. This is top-down processing, and it’s central to understanding why different people see different things in the same ambiguous image.

There’s one further detail worth knowing: the brain prioritises threatening or negative faces in processing. This is called the Threat Superiority Effect. Research has shown that negative or threatening face-like configurations are detected more quickly than neutral ones, even in cluttered visual fields. In pareidolia, the same bias applies. A face that seems angry or menacing is picked out faster than a benign one.

Top-Down Processing: Expectation Shapes Perception

One of the most important things pareidolia reveals about the mind is the power of top-down processing.

Perception runs on two parallel tracks. Bottom-up processing starts with the senses: light enters the eye, receptor cells fire, signals travel to the brain and get interpreted. It’s data-driven.

Top-down processing starts somewhere else entirely: with knowledge, expectations, beliefs, and prior experience. It flows downward from higher cognitive regions and shapes what the sensory data is allowed to mean. It’s concept-driven.

Pareidolia shows that top-down processing is powerful enough to override, or at least heavily shape, what the visual system extracts from real-world input. Tell someone a face is hidden in an image and their visual system will organise the same physical data differently than it would without that instruction.

A now classic experiment demonstrated this with audio. Participants were played reversed audio recordings. When told nothing, they heard noise. When told a specific message was hidden in the recording and shown the supposed lyrics, they heard the message clearly. Same stimulus. Different expectation. Completely different perceptual outcome.

This is the mechanism behind most religious pareidolia. Someone whose mental world is saturated with a particular face, whether a saint, a prophet, or a religious symbol, will find their perceptual system primed to extract that pattern from ambiguous visual input. They’re not lying. They’re not deluded. Their brain is doing what brains do: finding what it was set up to find.

Gestalt Psychology and Pareidolia: The Mind Completes the Picture

If you have any familiarity with Gestalt psychology, pareidolia is one of its clearest demonstrations. The Gestalt psychologists, principally Max Wertheimer, Kurt Koffka, and Wolfgang Köhler, working in the early twentieth century, argued that perception is holistic. The mind automatically organises parts into a coherent whole. Their foundational claim was that the whole is something other than the sum of its parts.

They identified several organising principles, all of which show up directly in pareidolia.

Closure is the tendency to complete incomplete shapes. When presented with a partial pattern, the mind fills the gaps. In pareidolia, a few dots or patches of shadow become a complete face because the brain closes the figure.

Similarity groups elements that share properties. Two dark patches in a rock face get grouped as “eyes” because they resemble each other and match an existing template.

Proximity links elements that are close together. A cluster of marks in a roughly face-like arrangement gets treated as a unified figure rather than unrelated points.

Figure and ground is perhaps the most famous Gestalt principle for pareidolia. The mind always assigns one element as the foreground figure and another as the background. The classic image that reads either as a white vase or as two black faces in profile is the standard illustration. Change which part you treat as figure and which as ground, and the entire percept flips.

Prägnanz, sometimes called the law of good form, holds that the mind tends to perceive stimuli in the simplest, most stable form available. A face is one of the most deeply familiar templates in the human brain. When incoming visual data is noisy or ambiguous, the mind defaults to its most established patterns.

Pareidolia is the result of all these principles running simultaneously on ambiguous input. The brain isn’t passively recording what’s there. It’s actively constructing a coherent percept from incomplete information, and it has very strong ideas about what that percept should look like.

Religion, Belief, and Who Sees What

Research published from the University of Helsinki and subsequently replicated at the University of Amsterdam found that religious individuals were more likely than non-believers to perceive faces and religious imagery in ambiguous visual stimuli. The effect held across both studies.

This finding is regularly misread. It doesn’t suggest that religious people have weaker perceptual systems or are more susceptible to error in some general sense. What it demonstrates is that prior beliefs and expectations feed directly into perception. This mechanism is universal. The content differs.

The person whose mental world contains strong, frequently activated representations of particular faces or symbols will find their visual system primed to extract those patterns when the input is ambiguous enough to permit multiple interpretations. They’re not seeing something others could also objectively see if they looked harder. They’re exporting internal content onto an external surface.

This is why, when face-like patterns appear in natural phenomena, people tend to see the figures most central to their own meaning-making systems. The face in the tree or the figure in the clouds reflects the viewer as much as it reflects the tree or the clouds.

The same logic applies outside religion. Sports fans have reported seeing team logos in cloud formations. Political partisans find symbolic patterns in random arrangements. The stimulus provides the raw material. The belief system does the identifying.

Pareidolia in Art History: Deliberate Use

Artists have been exploiting this perceptual tendency for centuries, and some of the most celebrated examples are entirely intentional.

Giuseppe Arcimboldo, the sixteenth-century Italian court painter, built an entire body of work around it. His portrait series depicted human faces assembled entirely from fruits, vegetables, flowers, and books. The “Librarian”, now held at Skansenmuseet in Sweden, shows a figure whose body is constructed from stacked volumes. Arcimboldo understood that the viewer’s brain would automatically resolve the arrangement of objects into a face, doing the perceptual work without being asked.

Salvador Dalí took this further and made it explicit. His painting “Slave Market with the Disappearing Bust of Voltaire” contains two entirely different images simultaneously. Depending on how the viewer approaches it, they see either a slave market scene or a bust of Voltaire. Dalí described this as his “paranoiac-critical method”: the deliberate use of ambiguous visual information to trigger competing perceptual interpretations. He understood the mechanism well before neuroscience had the tools to describe it.

In architecture, the principle appears in anthropomorphic building design. Façades with window arrangements that suggest eyes, doorways that read as mouths, and rooflines that imply a brow have been used since antiquity to make structures feel inhabited or watchful. The tendency to find humanoid features in built forms is so reliable that designers exploit it systematically.

In industrial design, research has shown that cars with “facial” front ends, headlights positioned like eyes, a grille that reads as a mouth, are perceived as more approachable and generate stronger emotional connection with users. Car designers are, at least informally, pareidolia engineers.

The Mars Face: When Science and Pareidolia Collide

The Mars face case is one of the most thoroughly documented pareidolia episodes in modern history, and it illustrates exactly how the phenomenon interacts with motivated reasoning.

In 1976, Viking 1 photographed a region of Mars called Cydonia. One image showed a mesa formation that, in the low-resolution photograph with particular lighting and shadow conditions, bore a strong resemblance to a humanoid face. NASA noted it in passing as an interesting example of shadows creating face-like formations. The press called it “the Face on Mars”. A subset of commentators argued it was evidence of an ancient extraterrestrial civilisation.

NASA’s position, that it was almost certainly a natural geological formation photographed under unusual conditions, was consistent but didn’t satisfy everyone. In 1998, the Mars Global Surveyor returned images of the same formation at significantly higher resolution. The face was gone. A mesa, ordinary and unremarkable, was visible in its place. A 2001 follow-up with even better resolution confirmed the same.

The explanation was rejected by some who had committed to the face interpretation. This is a predictable outcome, not a puzzling one. The face had never been the actual object of belief. It had been evidence recruited in support of a pre-existing belief in extraterrestrial civilisations. When the evidence dissolved, the belief it had been drafted to support remained intact.

Mars continues to produce pareidolia episodes. Images from rovers and orbiters regularly generate claims of spoons, lizards, human figures, and constructed objects. Each high-resolution follow-up resolves the ambiguity. The cycle repeats because the expectations driving it don’t require new evidence to persist.

Auditory Pareidolia: The Ears Do It Too

Pareidolia is not confined to vision. The auditory equivalent is well documented and follows the same underlying logic.

Backmasking became a cultural controversy in the United States in the 1980s. The claim was that certain rock records contained satanic messages that could be heard when the audio was played in reverse. Congressional hearings were held. Parental advisory labels followed.

Controlled research addressed the question directly. When participants were played reversed recordings without any instruction, they heard noise. When they were played the same recordings with the supposed hidden message written out in front of them, they heard the message. The audio was identical in both conditions. The expectation was not.

This is auditory top-down processing operating exactly as visual top-down processing does in face pareidolia. The brain searches available sensory data for patterns matching its active predictions. When told what to expect, it finds it.

Electronic Voice Phenomenon (EVP) works on the same basis. Investigators of claimed paranormal audio play recordings of static or ambient noise and perceive voices, words, or messages within them. Multiple laboratory studies have shown that the content of what is heard correlates strongly with what listeners are told to expect before listening. There’s no mystery here. The mechanism is the same one that finds faces in clouds.

What makes auditory pareidolia particularly interesting is that language processing in the brain is, if anything, even more automated and pre-conscious than face processing. The auditory cortex is primed to extract phonemic patterns from noise at remarkable speeds. It evolved in an environment where missing a spoken word from a group member or misreading a vocalisation from a predator both carried real costs.

Pareidolia and the Rorschach Test

The Rorschach inkblot test, introduced by Swiss psychiatrist Hermann Rorschach in 1921, is the most clinically institutionalised application of the pareidolia mechanism.

The procedure is straightforward: ten symmetrical inkblots, ambiguous by design, are shown in sequence. The subject says what they see. The underlying assumption is that ambiguous stimuli invite projection, that what the mind constructs from unclear input reflects the content of the inner world. Responses are coded and interpreted for personality and psychological functioning.

The pareidolia link is direct. The inkblots are specifically designed to be uninterpretable in any definitive way, forcing the visual system to draw on internal templates. The face-finding and pattern-completing tendencies that drive pareidolia in everyday life are what the Rorschach is banking on.

The clinical validity of the Rorschach is, however, a genuinely contested issue. Several criticisms have become increasingly difficult to dismiss. The normative data Rorschach originally used was collected from small, unrepresentative samples. Interpretation of responses depends heavily on the examiner’s skill and is vulnerable to systematic bias. Even the Exner Comprehensive System, developed to standardise scoring, has faced criticism regarding inter-rater reliability and predictive validity in independent research.

The British Psychological Society and the American Psychological Association both advise caution in using Rorschach data in isolation, particularly in high-stakes contexts like legal or diagnostic proceedings. The test is best understood as one source of clinical hypothesis rather than a standalone diagnostic instrument.

What the Rorschach demonstrates conclusively is that the pareidolia mechanism is rich enough to be clinically interesting. The specific images a person constructs from inkblots are not random. They are shaped by current preoccupations, emotional states, and habitual ways of organising experience. That’s real information, even if the scoring systems for extracting it remain imperfect.

Psychoanalytic Parallels: Transference as Relational Pareidolia

There’s a less obvious but theoretically important parallel between pareidolia and a concept central to psychoanalytic therapy: transference.

Transference, as Freud described it, occurs when a patient unconsciously displaces feelings, expectations, and relational patterns from significant past figures (usually parents) onto the therapist. The therapist, who typically works to be relatively neutral and opaque, becomes a kind of ambiguous stimulus. Into that ambiguity, the patient projects the familiar.

The structural similarity to pareidolia is close. An ambiguous external object (the therapist’s neutrality) is perceived through the lens of prior internal templates (relational history). The result is a perception shaped as much by the observer’s inner world as by anything objectively present in the other person.

In contemporary relational psychotherapy, transference is not treated as a distortion to be corrected but as a source of information. What the patient perceives in the therapist, and how that perception forms and shifts, tells the therapist things about the patient’s world that the patient couldn’t easily report directly. The therapist becomes a kind of clinical inkblot.

This isn’t a metaphor draped loosely over different phenomena. The cognitive mechanism is genuinely shared: an ambiguous stimulus, strong prior templates, and perception shaped by internal content rather than purely by external data.

Pareidolia in Childhood: Development and Imagination

Children experience pareidolia with considerably more frequency and vividness than adults. Every cloud holds an animal. Every knot in a wooden floorboard has a face. Every stain on a ceiling tells a story.

Two factors drive this. First, children are still building their taxonomic understanding of the world. The boundary between “pattern that resembles X” and “actual X” is less firmly established, so face-like configurations are more readily accepted as faces without the corrective processing adults apply automatically. Second, children’s imaginative engagement with their environment is less constrained by learned categorical thinking.

This is developmentally useful. Finding faces in trees, animals in clouds, and figures in shadows develops pattern recognition, imaginative capacity, narrative ability, and, crucially, the mental flexibility to hold multiple interpretations of the same object simultaneously. Pareidolia in childhood is part of cognitive development, not a sign of confusion about reality.

As children develop and accumulate experience, explicit top-down correction becomes more reliable. The adult brain still runs the same face-detection algorithms, but it applies a second-pass filter that asks: does this actually make sense? Under normal conditions, this suppresses the perceptual report. Under conditions of fatigue, fear, heightened expectation, or strong prior belief, the second-pass filter weakens and pareidolia becomes more prominent again.

Is Pareidolia Ever a Clinical Concern?

The clear answer is that pareidolia itself is not a disorder. It is a baseline feature of normal human perception and occurs in essentially everyone.

It can, however, appear as a symptom within clinical contexts or be amplified by certain conditions.

In schizophrenia, pareidolia may combine with delusional interpretation. The person doesn’t just see a face in a cloud. They are certain the face is watching them personally, sending them a message, or confirming a belief about a conspiracy directed at them. The pareidolia provides the raw perceptual material. The delusional system gives it a meaning that isn’t subject to ordinary reality testing.

Research on pareidolia in schizophrenia has produced mixed findings. Some studies report increased pareidolia; others report reduced pareidolia, possibly as a consequence of broader impairment in social and face-processing systems. The picture isn’t clean enough to use pareidolia as a standalone diagnostic marker.

In bipolar disorder during manic episodes, and in states of prolonged sleep deprivation or certain substance effects, pareidolia tends to intensify. This reflects changes in the general arousal and activation of the brain rather than a specific perceptual disorder.

The clinical distinction that matters most is this: can the person identify the perception as a construction of their own mind? Can they hold it lightly as “that looks like a face” rather than “that is a face with intentions directed at me”? The capacity for this kind of reality testing is what separates ordinary pareidolia from something requiring clinical attention.

A Simple Test Worth Trying

Next time someone tells you they’ve seen a significant face or figure in a natural object, try this: before they describe what they’ve found, look at the image yourself. Note what you see, or don’t see. Then ask them to point it out.

In most cases, you’ll see nothing meaningful until they direct your attention. Once they do, and once they’ve named what it is, the pattern becomes visible. That sequence, ambiguity first, identification second, pattern visible only after, is the hallmark of pareidolia rather than any objective feature of the image.

This isn’t a gotcha. It’s diagnostic. It tells you that what’s being reported is a perception shaped by expectation and knowledge, not a feature that any impartial observer would independently identify. That’s important information about what kind of claim is being made.

The same logic applies to large-scale pareidolia claims: the Mars face, the Jesus toast, the Marmite jar lid that someone in 1999 claimed bore the image of Mother Teresa. In every case, the pattern became visible to others after it was named and pointed to. The naming did as much work as the image.

What Pareidolia Tells Us About Perception

Pareidolia is one of the cleaner demonstrations that perception is not a passive recording of external reality. It’s a construction.

The brain receives fragmented, noisy, incomplete sensory data and generates a coherent experience from it. To do that, it uses prior knowledge, current expectations, emotional state, and attention. The output feels like direct access to the world. It isn’t. It’s a very fast, very good reconstruction.

This has broader implications. If the face-finding system can be triggered by random arrangements of light and shadow, then other perceptual systems are equally constructive. The certainty we feel about what we see, or hear, or remember, isn’t a reliable index of accuracy. It reflects the confidence of the construction, not its correspondence to anything external.

Pareidolia is the brain working correctly. The pattern-finding machinery is doing exactly what it evolved to do. The problem, when there is one, isn’t the perception. It’s what we do with it afterwards. Seeing is one thing. What we decide the seeing means is entirely another.

Frequently Asked Questions

References

1. Hadjikhani, N., Kveraga, K., Naik, P., & Ahlfors, S. P. (2009). Early (M170) activation of face-specific cortex by face-like objects. Neuroreport, 20(4), 403–407. https://doi.org/10.1097/WNR.0b013e328325a8e1 ↩︎