The predatory larvae of the caddis Plectrocnemia conspersa (Curtis) cause significant prey depletion in a habitat in which prey are patchily distributed. Optimal foraging theory predicts that under these circumstances a predator should stay in any given patch until the prey capture rate there drops to a value equal to the average for the habitat as a whole. This was tested using a combination of field and laboratory data and the results were in broad agreement with the prediction. A second prediction is that the marginal capture rate should be higher in a habitat richer in prey and this was not supported. It is argued that by using a simple rule-of-thumb (constant giving-up-time) P. conspersa is able to approach the optimal solution for much of the time.