Past research using a deferred imitation task has shown that 6-month-olds remember a 3-part action sequence for only 1 day. The concept of a time window suggests that there is a limited period within which additional information can be integrated with a prior memory. Its width tracks the forgetting function of the memory. This study asked if retrieving the memory of the modeled actions at the end of the time window protracts its retention, if the type of retrieval (active or passive) differentially influences retention, and if the retrieval delay influences its specificity. In Experiment 1, 6-month-olds either imitated the modeled actions (active retrieval group) or merely watched them modeled again (passive retrieval group) 1 day after the original demonstration. Both groups showed deferred imitation after 10 days. In Experiment 2, 6-month-olds who repeatedly retrieved the memory at or near the end of the time window deferred imitation for 2.5 months. In Experiment 3, 6-month-olds spontaneously generalized imitation late in the time window after 1 prior retrieval, whether it was active or passive. These studies reveal that the retention benefit of multiple retrievals late in the time window is huge. Because most retrievals are undoubtedly latent, the contribution of repeated events to the growth of the knowledge base early in infancy has been greatly underestimated. (Contains 4 figures.)