Research on children's understanding of video has shown seeming contradictions. Fourteen-month-olds imitate actions seen on TV (Meltzoff, 1988) and 18-month-olds are reminded of an event by watching video (Sheffield & Hudson, 2003) but 24-month-olds fail at a video-mediated object-retrieval task requiring dual representational understanding (Troseth & DeLoache, 1998). Experiments 1, 2A, and 2B tested whether changes to the video object-retrieval task that require "imitation" instead of "retrieval" would make it easier. Experiment 1 compared replications of Troseth and DeLoache's (1998) "window retrieval" and "video retrieval" conditions as well as "window imitation" and "video imitation" conditions, in which children imitated the act of finding a toy, to chance performance. Twenty-four-month-olds performed worse in the imitation conditions. Follow-up experiments showed that 24-month-olds' performance improved when an imitation search task included a goal but not when a retrieval task included a goal. Children transferred success after goal-based imitation to object retrieval. Implications for dual representation and the role of competing representations in the object-retrieval task are discussed. (Contains 4 figures and 3 footnotes.)