This article reviews the nature and function of recasts, a well-documented way of responding to young children. The paper challenges the definition of recast and argues that it is too broad a category to be useful, either for theories of language development or for practice. In particular, various forms of recast have featured in intervention programmes for children with language difficulties, but with mixed success. This review provides a theoretically motivated account of just those recasts that are likely to benefit children with language difficulties. To this end, the Contrast theory of corrective input is invoked, where the focus is on adult models that are directly contingent on child errors (Saxton, 1997). Both theoretical and empirical evidence suggests that this kind of input can facilitate the acquisition of adult-like grammatical competence.