We present a neural network model of how events are stored in and retrieved from episodic long-term memory (LTM). The model is novel in giving an explicit account of the working memory (WM) medium mediating access to episodic memory: it makes a specific proposal about how representations of events and situations in semantic WM interface with representations of events and situations in episodic memory. It also provides the framework for an account of how operations accessing temporally remote situations are reported in language.