Episodic memory involves binding together what-where-when associations. In three experiments, we tested the development of memory for such contextual associations in a naturalistic setting. Children searched for toys in two rooms with two different experimenters; each room contained two identical sets of four containers, but arranged differently. A distinct toy was hidden in a distinct container in each room. In Experiment 1, which involved children between 15 and 26 months who were prompted with a very explicit cue (a part of the hidden toy), we found a marked shift in performance with age: while 15- to 20-month-olds concentrated their searches on the two containers that sometimes contained toys, they did not distinguish between them according to context, but 21-26-month-olds did. However, surprisingly, without toy cues, even the youngest children showed a fragile ability to disambiguate the two containers by room context. In Experiment 2, we tested 34- to 40-month-olds and 64- to 72-month-olds without toy cues. The 5-year-olds were nearly perfect, and the 3-year-olds showed a significant preference for the correct container given only the context. In Experiment 3, we filled in the age range, and also investigated the effects of the use of labels (i.e. names of experimenters and rooms) and of familiarization time, in groups of 34- to 40-month-olds, 42- to 48-month-olds, and 50- to 56-month-olds. Neither labels nor familiarization time had an effect. Across experiments, there was regular age-related improvement in context-based memory. Overall, the results suggest that children's episodic memory may undergo an early qualitative change, yet to be precisely characterized, and that continuing increments in the use of contextual cues occur throughout the preschool period. A video abstract of this article can be viewed at https://www.youtube.com/watch?v=DkwEFw0UEz4&list=PLwxXcOKHPC0llAPVcJyW4EtzlA934A2Rz&index=1