The ability to recognize one's own inner speech is essential for a sense of self. The verbal self-monitoring model proposes that this process entails a communication from neural regions involved in speech production to areas of speech perception. According to the model, if the expected verbal feedback matches the perceived feedback, then there would be no change in activation in the lateral temporal cortices. We investigated the neural correlates of verbal self-monitoring in a functional magnetic resonance (fMRI) study. Thirteen healthy male volunteers read aloud presented adjectives and heard their auditory feedback which was experimentally modified. Decisions about the source of the feedback were made with a button-press response. We used a 'clustered' fMRI acquisition sequence, consisting of periods of relative silence in which subjects could speak aloud and hear the feedback in the absence of scanner noise, and an event-related design which allowed separate analysis of trials associated with correct attributions and misattributions. Subjects made more misattribution responses when the feedback was a distorted version of their voice. This condition showed increased superior temporal activation relative to the conditions of hearing their own voice undistorted and hearing another person's voice. Furthermore, correct attributions during this condition were associated with greater temporal activation than misattributions. These findings support the self-monitoring model as mismatches between expected and actual auditory feedback were associated with greater temporal activation.