In two experiments, both employing deferred imitation, we studied the developmental origins of episodic memory in two- to three-year-old children by adopting a "minimalist" view of episodic memory based on its What-When-Where ("WWW": spatiotemporal plus semantic) content. We argued that the temporal element within spatiotemporal should be the order/simultaneity of the event elements, but that it is not clear whether the spatial content should be egocentric or allocentric. We also argued that episodic recollection should be configural (tending towards all-or-nothing recall of the WWW elements). Our first deferred imitation experiment, using a two-dimensional (2D) display, produced superior-to-chance performance after 2.5 years but no evidence of configural memory. Moreover, performance did not differ from that on a What-What-What control task. Our second deferred imitation study required the children to reproduce actions on an object in a room, thereby affording layout-based spatial cues. In this case, not only was there superior-to-chance performance after 2.5 years but memory was also configural at both ages. We discuss the importance of allocentric spatial cues in episodic recall in early proto-episodic memory and reflect on the possible role of hippocampal development in this process.