SUMMARY Memory systems have evolved under selection pressures, such as the need to remember the locations of resources or past events within spatiotemporally dynamic natural environments. The full repertoire of complex behaviours exhibited by animals in dynamic surroundings are, however, difficult to elicit within simply structured laboratory environments. We have developed a computer-controlled naturalistic environment with 64 feeders for simulating dynamic patterns of water or food resource availability (depletion and replenishment) within the laboratory. The combination of feeder and cage remote control permits the automated transfer of animals between cage and test arena and, therefore, high experimental throughput and minimal disturbance to the animals (bats and mice). In the present study, we investigated spatial working memory in nectar-feeding bats (Glossophaga soricina,Phyllostomidae) collecting food from a 64-feeder array. Feeders gave only single rewards within trials so that efficient foraging required bats to avoid depleted locations. Initially, bats tended to revisit feeders (win-stay), but within three trials changed towards a win-shift strategy. The significant avoidance of revisits could not be explained by algorithmic search guiding movement through the array nor by scent cues left by the bats themselves and,thus, the data suggest that bats remembered spatial locations depleted of food. An examination of the recency effect on spatial working memory after bats shifted to a win-shift strategy indicated that bats held more than 40 behaviour actions (feeder visits) in working memory without indication of decay. This result surpasses previous findings for other taxa.