The purpose of this study was to investigate the validity of an explicit generation task in order to assess the explicit knowledge acquired by the subjects in an artificial grammar learning task. To determine the sensitivity of the generation task and to ensure that the knowledge shown in this task is actually related to information learned during the study phase (and not to information learned during the classification phase), a group of normal subjects were administered the generation task in different conditions: after the classification phase with and without a study phase, or immediately after the study phase. The main result is that performance in the generation task is specifically related to information learned during the study phase and not to information presented during the classification phase; moreover, there was no correlation between the implicit and explicit measures, suggesting that the generation task can be considered as a valid measure of explicit knowledge in artificial grammar learning tasks.