There is broad agreement that context plays a role in episodic memory. There is less agreement on the nature and centrality of this role. Retrieved context models specify a set of computational mechanisms that place context at the very heart of episodic memory processes. They assume that an internal context representation updates, or drifts, whenever new events are experienced and that these events form associations to this representation. They further assume that memory search proceeds by using the context representation as a cue and that recalled memories reinstate their own context which then forms part of the cue for the next recall. When applied to laboratory tasks such as free recall, these models make specific testable predictions about not only recall behavior but also to the neurophysiology of memory search. Specifically, they predict that the brain should maintain a representation that changes gradually during study and that new events should become associated with this context representation. Most critically, they predict that during memory search successfully recalling one item should trigger reinstatement of the state of context associated with that item. If true, this should manifest in subjects’ behavior as a tendency for items experienced nearby in time to cluster together during recall (i.e., a contiguity effect) and in neurophysiological recordings as a temporally graded pattern of reinstatement. We review the evidence for each of these predictions. We also discuss a range of related issues including group and individual variation in contextual processing and different types of context, such as temporal and source.