Researchers have claimed that recasts might be ambiguous as feedback. Because recasts serve a dual function, as both feedback and conversational response, learners might not always interpret them as feedback (e.g., Lyster & Ranta, 1997). This study explores how learners interpret recasts they notice (as responses to content, negative evidence, positive evidence, or a combination of negative and positive evidence) and how recast features (linguistic targets, length, number of changes) might affect their interpretations. Forty-nine learners of Japanese engaged in task-based activities during which they received recasts of morphosyntactic and lexical errors. When learners noticed recasts, they occasionally interpreted them as responses to content, particularly when recasts were long and substantially different from their problematic utterances. In contrast, learners were significantly more likely to attend to the linguistic evidence in recasts when these were short and closely resembled the original utterances. These patterns were generally observed for both morphosyntactic and lexical recasts. Results suggest that length and number of changes might, in part, determine the explicitness of recasts as feedback and thus affect learners' abilities to interpret them as such.