According to the discrepancy-attribution hypothesis (B. W. A. Whittlesea & L. D. Williams, 1998), people experience a feeling of familiarity when they perceive their processing to be surprising, but for an indefinite reason. This hypothesis has been successful in explaining several illusions of familiarity. Here, it is applied to the prototype-familiarity effect, an illusion of remembering that occurs when people are shown prototype words after studying lists of associates. The experiments showed that studying associates enhances semantic, but not perceptual, processing of prototypes. They also showed that claims of recognizing prototypes can be modified by presenting them in predictive or incongruous contexts at test. The evidence suggests that the effect results from an evaluation process that monitors the coherence of processing.