How do people infer others' perceptions when stimuli afford multiple interpretations? We tested whether perceptual reversal (reporting multiple readings of ambiguous figures) supports Theory of Mind (ToM) and reduces self-projection.


Japanese adults (N = 300; aged 20-70) viewed three bistable figures: Young woman/Old lady, Duck/Rabbit, Man/Rat. After a 5-second preview with replay available, participants completed: (a) untimed free-text "everything you saw," (b) initial percept (forced choice), and (c) what third persons (aged 3/20/75) would perceive (side-1/side-2/cannot determine). Reversal was coded when both interpretations appeared in free text. ToM accuracy, following Carpendale & Chandler (1996), required selecting "cannot determine" as correct. Self-projection was coded when incorrect responses matched own initial percepts.


Binomial GEE models (clustered by participant) included Figure × Target Age, Reversal, demographics, and interactions. Despite unlimited viewing time, reversal remained uncommon: Young woman/Old lady (7%), Duck/Rabbit (23%), Man/Rat (13%).


Model 1: Reversal predicted higher ToM accuracy (OR = 1.54, 95% CI [0.98, 2.42]), improving accuracy by 4-11 percentage points (Duck/Rabbit: +7; Man/Rat: +4-5; Young woman/Old lady: +11). Young woman/Old lady exceeded Duck/Rabbit for ToM accuracy (OR = 1.40 [1.10, 1.77]), except when judging 75-year-old targets (interaction OR = 0.67 [0.49, 0.91]). Older participants endorsed indeterminacy less (60s OR = 0.57; 70s OR = 0.52) than younger people.


Model 2: Among incorrect responses, reversal reduced projection (OR = 0.54 [0.37, 0.78]), decreasing projected errors by 7-14 points. Projection increased for Young woman/Old lady with 20-year-old targets (interaction OR = 2.11 [1.07, 4.15]) and with participant age (60s OR = 1.80; 70s OR = 2.18).


Perceptual flexibility supports ToM-appropriate uncertainty and buffers against egocentric biases. Experiencing multiple visual interpretations helps people recognize that others face similar ambiguity, rather than projecting their own percepts. Limitations include a Japanese-only sample and conservative reversal coding.