Childers and Tomasello (2001) found that training 21/2-year-olds on the English transitive construction greatly improves their performance on a post-test in which they must use novel verbs in that construction. In the current study, we replicated Childers and Tomasello's finding, but using a much lower frequency of transitive verbs and models in training. We also used novel verbs that were of a different semantic class to our training verbs, demonstrating that semantic homogeneity is not crucial for generalization. We also replicated the finding that 4-year-olds are significantly more productive than 21/2-year-olds with the transitive construction, with the new finding that this is also true for verbs of emission. In addition, ‘shared syntactic distribution’ of novel verb and training verbs was found to have no observable effect on the number of 21/2-year-olds who were productive in the post-test.