There is growing evidence that corrective input for grammatical errors is widely available to children (Farrar, 1992; Morgan, Bonamo & Travis, 1995). However, controversy still exists concerning the extent to which children can identify and exploit available negative input. In particular, very little is yet known about the longer-term effects of negative input. Performing a time series analysis on observational data, Morgan et al. (1995) conclude that corrective recasts are not related to future improvements in grammaticality. It is argued here, though, that the data sets analysed in this study are inherently ill-suited to the demands of time series analyses. The present study adopts an experimental approach in order to compare the effects of negative evidence versus positive input on the acquisition of irregular past tense verb forms. Twenty-six children (mean age 3;10) participated in a within-subjects design over a period of five weeks. It was found that improvements in the grammaticality of child speech were considerably greater in cases where negative evidence had been provided. Moreover, children's intuitions concerning the status of irregular and overregularized forms more closely approximated adult intuitions when corrective input was available.