Using traditional implicit and explicit artificial-grammar learning tasks, the author investigated the similarities and differences between the acquisition of declarative knowledge under implicit and explicit learning conditions and the functions of the declarative knowledge during testing. Results suggested that declarative knowledge was not predictive of or required for implicit learning but was related to consistency in implicit learning performance. In contrast, declarative knowledge was predictive of and required for explicit learning and was related to consistency in performance. For explicit learning, the declarative knowledge functioned as a guide for other behavior. In contrast, for implicit learning, the declarative knowledge did not serve as a guide for behavior but was instead a post hoc description of the most commonly seen stimuli.