A number of interaction researchers have claimed that recasts might be ambiguous to learners; that is, instead of perceiving recasts as containing corrective feedback, learners might see them simply as literal or semantic repetitions without any corrective element (Long, in press; Lyster & Ranta, 1997). This study investigates learners' interpretations of recasts in interaction. Videotapes of task-based interactions including recasts and repetitions were shown to advanced English as a second language students (N = 34). Although both groups viewed the teacher's feedback (recasts, repetitions, or other), one group saw video clips that had been edited to remove the learners' nontargetlike utterances that had triggered the feedback, and another group saw the same video clips with the initial nontargetlike utterances included. After each clip, learners in both groups were asked to indicate whether they thought they were hearing a recast, a repetition, or something else. A subset of learners (n = 14) provided verbal reports while they evaluated the clips. Results show that learners who did not overhear initial learner utterances were significantly less successful at distinguishing recasts from repetitions. The verbal protocol data suggest that learners were not looking for nonverbal cues from the speakers. A post hoc analysis suggests that morphosyntactic recasts were less accurately recognized than phonological or lexical recasts in this study. These findings suggest that the contrast between a problematic utterance and a recast contributes to learners' interpretations of recasts as corrective.