Constellation · a figure drawn on static
Pareidolia
Pareidolia is the spontaneous perception of a meaningful, familiar pattern - most often a face - in vague, ambiguous, or random visual or auditory input that does not actually contain it.
The dots are scattered at random. The figure is yours.
How the figure gets drawn
The dominant account is a broadly tuned, high-sensitivity face-detection system that privileges sensitivity over selectivity: minimal face-like configurations (two 'eyes' above a 'mouth') trigger face-specific cortex (right FFA) early (~165-170 ms), the same timing as real faces (Hadjikhani 2009). This bottom-up template is modulated top-down by expectation, with frontal regions (inferior frontal gyrus) biasing visual cortex toward 'seeing' a face (Liu 2014). Evolutionarily, a system that occasionally false-alarms on faces/agents is favored over one that misses real faces or threats - a 'better safe than sorry' detector, linked to hyperactive agency detection. Wardle (2022) extends this: detection alone is sufficient to read a face as male, so the default 'face' template carries social biases.
Competing account — Debate persists over how purely bottom-up the effect is: Hadjikhani frames early fusiform activation as evidence against late cognitive reinterpretation, whereas Liu emphasizes top-down frontal expectation as causally necessary. There is also dispute about how far pareidolia faces engage downstream social/attentional processing: Palmer & Clifford (2020) and some prioritization studies argue illusory faces recruit social-attention mechanisms and attract attention, while Miti et al. (2025) failed to replicate attentional capture, suggesting the shared machinery may be limited to early detection rather than automatic attentional prioritization.
Patterns people swore they saw
Objects incidentally perceived as faces evoked an early ~M165/M170 activation in ventral fusiform cortex (FFA), at timing/location similar to real faces; common objects did not. Authors concluded face perception of face-like objects is an early process, not a late cognitive reinterpretation.
Nouchine Hadjikhani, Kestutis Kveraga, Paulami Naik, Seppo P. Ahlfors · 2009 · not reported in retrieved abstract
Reporting an illusory face activated the right fusiform face area, and right-FFA response correlated with susceptibility to pareidolia; expectation signals from frontal cortex (inferior frontal gyrus) helped drive the illusory percept, supporting a top-down + bottom-up account.
Jiangang Liu, Jun Li, Lu Feng, Ling Li, Jie Tian, Kang Lee · 2014 · n=20 (11 males, ages 18-25, right-handed, healthy Chinese adults)
Illusory faces are reliably attributed emotion, age, and gender; a strong male bias emerged (~4:1). 90% of images had a male mean rating vs 9% female; 81.4% of all gender ratings were male; 80% of participants were male-biased vs 3% female-biased. Bias not explained by object semantics or visual features; human face morphs instead skewed female.
Susan G. Wardle, Sanika Paranjape, Jessica Taubert, Chris I. Baker · 2022 · n_total = 3,815 adults across experiments (Exp 1a: n = 800, 100 ratings per image)
Adapting to pareidolia faces with an apparent attentional direction systematically biased perception of where real human faces were looking, indicating shared higher-level coding of social attention (not just low-level features) between illusory and real faces.
Colin J. Palmer, Colin W. G. Clifford · 2020 · not reported in retrieved summary
Evidence — mixed. The core phenomenon is robust and well replicated: illusory faces engage face-selective cortex (right FFA) at face-like latencies, and the male-gender bias is highly reliable (n=3,815; split-half rs=0.85; replicated across experiments and rater gender, and found in children). However, specific downstream claims are contested: Miti, Ciaramidaro, Rubichi & Iani (2025) report a direct failure to replicate attentional capture by task-irrelevant real OR illusory faces, finding instead that illusory faces gave a search DISADVANTAGE when task-relevant. So 'faces (real and illusory) automatically grab attention' should not be treated as settled, even though early-detection and gender-bias findings are strong.
Figures seen in the field
Aerospace / planetary science · 1976
The 'Face on Mars' (Cydonia)
On 25 July 1976, NASA's Viking 1 orbiter imaged a mesa in the Cydonia region (frame 035A72) that, under ~10-degree low-angle lighting, resembled a humanoid face. It became a decades-long touchstone for claims of a Martian civilization. Higher-resolution imagery from NASA's Mars Orbiter Camera (2001), mounted on Mars Global Surveyor, showed an ordinary eroded mesa, making it the textbook case of cosmic pareidolia.
Media / audio · 1969
'Paul is dead' backmasking in Beatles recordings
From September 1969, listeners claimed hidden clues to Paul McCartney's supposed death were encoded in Beatles records, including phrases allegedly heard when tracks were played backward. Such backmasking claims are a documented form of auditory pareidolia - perceiving meaningful speech in audio not designed to contain it.
Aerospace / planetary science · 2023
'Teddy bear', 'shark fin', and 'crab claw' on Mars (MRO/HiRISE and Perseverance imagery)
In December 2022 (viral January 2023), NASA's Mars Reconnaissance Orbiter HiRISE camera photographed surface features widely reported to resemble a teddy bear's face - an impact crater forming the nose, two craters for eyes, and a circular hill as the head. Separately, in August 2023, NASA's Perseverance rover photographed rock formations that amateur observers identified as a 'shark fin' and a 'crab claw.' These are contemporary, independently dated instances of planetary pareidolia driving public misinterpretation of Mars imagery.
To tell signal from noise
Increase resolution and seek independent, unprimed observers: get better data (higher-res image, cleaner audio, larger sample) and ask whether people who weren't told what to look for perceive the same pattern. For signals/data, pre-register what would count as a real pattern and test against a noise/null model before interpreting; treat a vivid percept in ambiguous input as a hypothesis, not evidence.
The 'Face on Mars' dissolved once NASA's Mars Orbiter Camera supplied higher-resolution imagery (~2001), and Liu et al. (2014) showed that mere expectation/priming from frontal cortex can manufacture an illusory-face percept in pure noise - so removing priming and improving data directly attack the mechanism.
Catch it in the act
You're confidently 'seeing' a specific face, figure, or message in something that is actually noise, texture, or random structure - a cloud, a rock, an electrical outlet, static, or audio played backward - and the percept feels immediate and obvious rather than inferred. Watch for it especially when the input is low-resolution or ambiguous and you've been primed to expect a particular pattern. A tell is that other people, unprimed, often don't see it, and higher-resolution data makes it dissolve.
Related pattern errors
- ApopheniaparentCoined by Klaus Conrad (1958) for the 'unmotivated seeing of connections' with a sense of abnormal meaningfulness across any data. Pareidolia is the perceptual (chiefly visual/auditory) subtype.
- PatternicitysiblingMichael Shermer's term for the tendency to find meaningful patterns in meaningless noise; overlaps heavily with apophenia/pareidolia as a framing.
- Hyperactive agency detectionmechanistically-linkedProposed evolutionary mechanism (better to false-alarm on an agent/face than miss a predator) often invoked to explain why face pareidolia is adaptive.
- Illusory correlationsiblingPerceiving covariation between unrelated variables; a non-perceptual cousin within the broader 'seeing structure in noise' family (Not Enough Meaning quadrant).
- Confirmation biaseasily-confusedOften co-listed but distinct: pareidolia is a low-level perceptual illusion, whereas confirmation bias is a reasoning/evidence-selection bias.
- Clustering illusionsiblingSeeing 'clusters' or streaks in random sequences; the statistical analogue of perceptual pareidolia.