We tested the predictions of 2 explanations for retrieval-based learning; while the elaborative retrieval hypothesis assumes that the retrieval of studied information promotes the generation of semantically related information, which aids in later retrieval (Carpenter, 2009), the episodic context account proposed by Karpicke, Lehman, and Aue (in press) assumes that retrieval alters the representation of episodic context and improves one's ability to guide memory search on future tests. Subjects studied multiple word lists and either recalled each list (retrieval practice), did a math task (control), or generated associates for each word (elaboration) after each list. After studying the last list, all subjects recalled the list and, after a 5-min delay, recalled all lists. Analyses of correct recall, intrusions, response times, and temporal clustering dissociate retrieval practice from elaboration, supporting the episodic context account.