Narrative descriptions of events often depart from how these events would have occurred in "real time." For example, narratives often contain time shifts in which events that are irrelevant to the plot are omitted. Zwaan (1996) has shown that these time shifts may affect on-line comprehension. Specifically, they are associated with increases in processing load and a deactivation of previous information. The experiments in the present article show that the situation is more complex. Specifically, there is only a deactivation of previous events if they are not assumed to be ongoing after a time shift. Furthermore, explicit discontinuations of events, as in he stopped walking also lead to deactivations when compared with explicit continuations and resumptions.The general assumption in research on narrative comprehension is that comprehenders are influenced by the nature of the situation that is being described. The idea is that comprehenders place themselves, to some extent, "in" the narrated situation. Such a shift from an actual situation to a described situation has been termed a deictic shift (Duchan, Bruder, & Hewitt, 1995; see also MacWhinney, 1999). As a consequence of this deictic shift, people, objects, and events that are in the narrated situation should be more accessible to comprehenders than people, objects, and events that are not. That is, nodes that code for these situational entities should be more activated when the entities are present in the situation than when they are not.Zwaan (1999b) has reviewed various findings in the literature that are consistent with this idea. Objects that are in the situation yield shorter recognition latencies to their associated words than do objects that are not in the situation. For instance, response latencies to the word cake are faster when the text states that Mary baked cookies and cake than when the text states that Mary baked cookies but no cake, despite the fact that the word occurs in both sentences (Macdonald & Just, 1989). This is not due simply to the presence of the word not. Kaup (1999) obtained similar effects for He destroyed the desk, but not the table. In this case, the object qualified by not was in the situation and its associated word yielded faster recognition responses. Similarly, a memory node that codes forWe thank Eileen Goodwin, Loretta Handegard, LaRicka Wingate, and Daisy Winkler for their assistance with data collection and Mike Rinck and Tom Trabasso for their helpful comments on a previous version. Address all correspondence to R. A. Zwaan,