The ability to categorize objects and events in the world around us is a fundamental and critical aspect of human learning. We trained healthy adults on a probabilistic category-learning task in two different training modes. The aim of this study was to see whether either form of probabilistic category learning (feedback or observational) undergoes subsequent enhancement during sleep. Our results suggest that after training, a good night of sleep can lead to improved performance the following day on such tasks.[Supplemental material is available online at http://www.learnmem.org.]From infancy, humans categorize objects and events they encounter. In our daily lives, this category learning enables us to extract rules and patterns from our varied experiences, and to formulate new responses and inferences to familiar objects and events, thereby facilitating decision-making and problem-solving processes .While there is now considerable understanding of the early stages of category learning, relatively little is known about how such learning evolves over time. A growing literature shows that post-training sleep plays an important role in long-term memory consolidation and enhancement (Stickgold 2005;Diekelmann et al. 2009). The most compelling evidence thus far comes from simple perceptual and motor procedural learning, such as fingertapping sequences and visual discrimination tasks (Karni et al. 1994;Gais et al. 2000;Stickgold et al. 2000;Walker et al. 2002;Huber et al. 2004). Much less is understood about the role of sleep in the learning of explicit episodic memory tasks, and even less for more complex memory tasks, such as probabilistic category learning (Smith and Smith 2003;Wagner et al. 2004;Ellenbogen et al. 2007).There are two basic patterns by which categorizations can be learned: We can simply observe stimuli and their outcomes (observational learning), or we can be shown stimuli and then asked to predict outcomes, only then receiving feedback on the accuracy of our response (feedback learning). The Weather Prediction Task (WPT) was initially described by Knowlton et al. (1994) and has been extensively studied as a paradigm for such category learning (Knowlton et al. 1994;Gluck et al. 2002;Meeter et al. 2008).Studies have shown that the WPT activates different memory systems depending on how the task is learned and on the extent of training. Neuroimaging studies have shown that subjects who simply observe stimuli and their outcomes activate learning and memory systems that are supported by the medial temporal lobe (MTL) and prefrontal cortex (observational learning, Fig. 1, top). On the other hand, subjects who must predict outcomes for each stimulus and only then receive feedback on their accuracy (feedback learning, Fig. 1, bottom), initially activate the MTL, but as training continues shift this activation to the striatum, which is known to be involved with habit and skill-learning behaviors (Poldrack et al. 2001).Evidence from studies with patient populations supports this conclusion. Patients in whom b...