Auditory verbal hallucinations (AVHs) are typically associated with schizophrenia but also occur in individuals without any need for care (nonclinical voice hearers [NCVHs]). Cognitive models of AVHs posit potential biases in source monitoring, top-down processes, or a failure to inhibit intrusive memories. However, research across clinical/nonclinical groups is limited, and the extent to which there may be continuity in cognitive mechanism across groups, as predicted by the psychosis-continuum hypothesis, is unclear. We report two studies in which voice hearers with psychosis (n = 31) and NCVH participants reporting regular spiritual voices (n = 26) completed a battery of cognitive tasks. Compared with non-voice-hearing groups (ns = 33 and 28), voice hearers with psychosis showed atypical performance on signal detection, dichotic listening, and memory-inhibition tasks but intact performance on the source-monitoring task. NCVH participants, however, showed only atypical signal detection, which suggests differences between clinical and nonclinical voice hearers potentially related to attentional control and inhibition. These findings suggest that at the level of cognition, continuum models of hallucinations may need to take into account continuity but also discontinuity between clinical and nonclinical groups.