A number of descriptive studies of language classrooms have identified recasts as a frequent form of feedback used by teachers following learners' nontarget-like oral production. Some classroom-based researchers (e.g., Lyster, 1998) have suggested that recasts are less effective than other forms of feedback because of the ambiguity of their potentially corrective purpose. The present study focused on both the provision and the effectiveness of recasts in 12 adult English second language classrooms throughout 17 hours of meaning-based interaction. There were 12 teachers and 118 learners who participated, with class sizes ranging from 6 to 14 students. Comparisons involving the incidence of recasts, elicitation, and metalinguistic feedback, together with learner responses (e.g., successful uptake) following these types of feedback, revealed that recasts were widely used and, similar to other types of corrective feedback, were beneficial at least 50% of the time, as measured by posttests. The recasts differed according to characteristics that emphasized their corrective purpose. Logistic regression analysis revealed certain characteristics that were associated with successful uptake and with accuracy on posttests. Stress, declarative intonation, one change, and multiple feedback moves were predictive of successful uptake, whereas interrogative intonation, shortened length, and one change were predictive of the accuracy of the test scores. This study suggests that recasts vary in implicitness and that these differences may have an impact on their effectiveness, both in terms of learners' successful uptake and subsequent use. Moreover, the ambiguity of recasts is greatly reduced by the phrasal, prosodic, and discoursal cues that teachers provide. The effectiveness of recasts is likely to be affected by these cues and other factors, such as degree of difference between the recast and the nontarget-like utterance.