This study investigated the interactions between feedback type, proficiency, and the nature of the linguistic target in the learning of Chinese as a foreign language. Seventy-eight learners from two large US universities participated in the study. The participants were divided into two proficiency levels based on their performance on a standardized proficiency test. At each proficiency level, they were randomly assigned to three feedback conditions: recasts, metalinguistic correction, and control. Learners in the experimental conditions received feedback on their nontargetlike use of classifiers and the perfective "-le." Results revealed that for the perfective "-le," recasts benefited the high-level but not low-level learners; at the high proficiency level, the effects of recasts were more sustainable than those of metalinguistic correction. With respect to classifiers, recasts were effective for learners at both proficiency levels. For both target structures, metalinguistic correction showed larger effects than recasts for the low-level learners, but the two feedback types were equally effective for the advanced learners. The results underscore the importance of taking an interactional approach to the investigation of corrective feedback. The results also undermine the commonly believed superiority of explicit feedback over implicit feedback.