Why does it take less time to remember an event than to experience it? Recent evidence suggests that the dynamic unfolding of events is temporally compressed in memory representations, but the exact nature of this compression mechanism remains unclear. The present study tested two possible mechanisms. First, it could be that memories compress the course of events into a sequence of moments or slices of prior experience, while omitting other segments, akin to edited films that give condensed accounts of events using sequences of separate shots (referred to as the "discontinuity hypothesis"). Alternatively, it may be that the entire stream of information is represented but is mentally replayed at a faster speed than the original experience (referred to as the "acceleration hypothesis"). In two experiments, these hypotheses were tested by comparing mental replay times for continuous movies depicting naturalistic events and edited versions of the same movies in which less informative parts were removed to mimic the presumed structure of memory representations according to the discontinuity hypothesis. We found that memories for videos in which less informative segments were replaced by temporal ellipses (Experiment 1) or by black screens of the same duration as removed segments (Experiment 2) were less compressed and contained a higher density of recalled units than did memories for complete videos. These results support the discontinuity hypothesis and suggest that segments of time that are redundant and predictable are omitted in episodic memory, while more informative segments are selectively retained to represent the unfolding of events.