Lexical retrieval is an essential subprocess in language production, and its efficiency is crucial for writing. To improve writing quality in a second language, we developed an experimental, computerized training for improving fluency of lexical retrieval in a classroom setting, applying techniques previously restricted to laboratory use. In a counterbalanced design with randomized assignment, each of two groups was trained on a different set of words for productive use. A previous study showed that students in both groups attained greater fluency of lexical retrieval on the trained word set in comparison to students who were not trained on that word set. The current study provides evidence that this enhanced fluency transferred to narrative writing, as students in both groups used the trained words more often in narrative texts. In addition, one of the groups showed significant improvement in their expression of essential content elements. However, no significant differences were found on global quality ratings. The results are discussed in the context of theories of limited processing capacity during text production. We also consider the implications of these results for language instruction.