<p>Narrative time is essential for understanding and remembering stories. Reconstructive memory theory posits that retrieving past events is not a mere reactivation of the original memory trace but involves a reorganization process informed by a combination of stored memories, general knowledge, and interpretative elements. Recent studies have shown that humans are remarkably accurate in judging the time-of-occurrence of fragments from a previously encoded narrative, but also that expectations and assumptions can compromise their performance. Here, we investigated the mechanism underlying the ability to infer the time-of-occurrence of videoclips extracted from a previously unencoded movie to elucidate the gradual integration of episodic and semantic information during the construction of narrative time. Across four experiments performed by different groups of human participants, we progressively manipulated the amount of available episodic information for the time-estimation task. Compared with the high precision observed for a known (i.e., previously encoded) movie, a robust decrease in performance was observed in the absence of prior encoding, irrespective of task repetitions. Exposing participants to additional episodic information (movie fragments) between task repetitions produced a gradual enhancement in task performance, especially when episodic cues were presented in chronological order. These results suggest that the temporal information provided by episodic cues can be exploited to gradually form a temporal scaffolding of the narrative, filling in the gaps between encoded pieces of information. This temporal representation, in turn, enables the dating of movie fragments, almost as if the movie had been encoded in its entirety.</p>