Hallucinations are characterized by disturbances of perceptual processes involved in decision-making about environmental stimuli. Here, we examine whether cognitive and computational processes by which sensory information is integrated may offer insight into the perceptual mechanisms of hallucinatory symptoms. We used a multi-element perceptual averaging task in which observers made dichotomous judgments about the “average color” (red or blue) of an array of stimuli in trials that varied in the strength (mean) and reliability (variance) of the decision-relevant perceptual evidence. Generally, observers excluded or down-weighted extreme (outlying) perceptual evidence akin to a statistician excluding outlying data points; however, individuals prone to hallucinations afforded more weight to more extreme or untrustworthy evidence. Computational modeling showed that individuals prone to hallucinations tended not to use the optimal model in which evidence is integrated as a function of the log odds of each perceptual option leading to “robust averaging”. Finally, observers generally adapted to trials with unreliable evidence by increasingly downweighting extreme evidence, but the weighting strategy in hallucination prone individuals remained insensitive to the reliability of evidence. By showing that hallucination proneness is associated with reduced attenuation of untrustworthy evidence in perceptual decision-making, our findings suggest a novel perceptual mechanism underlying hallucinations. Our findings also provide support for the view that hallucination-proneness relates to alterations in the perceptual systems that track statistical regularities in environmental stimuli.