Purpose: Although conversational recasting has been a generally successful treatment approach, the precise factors that influence children's learning through recasts are not yet understood. In this study, we examined details of the relationship between child utterance and clinician utterance that seemed likely to influence learning. Method: Three measures were calculated from transcripts of recasting sessions with 17 preschoolers with specific language impairment. In all sessions, 3rd person singular -s served as the target. The measures of interest were the frequency of recasts following child utterances that were prompted by clinicians, the frequency of clinicians' recasts of subjectless sentences, and the frequency of clinicians' noncorrective recasts. We assessed the short-term and long-term predictive value of these measures through regression analyses. Results: Noncorrective recasts proved to be a positive predictor of short- and long-term gains in the use of the target form. Recasts of subjectless sentences were associated with poorer outcomes, though their contribution was relatively small. Conclusions: The nature of learning that takes place varies according to the relationship between child and clinician utterances during the recasting process. These variations have implications for clinical practice and for how learning through recasting is characterized. (Contains 4 tables.)