Abstract: An experiment with starlings was conducted to investigate the effect of variability in inter-reinforcement intervals on foraging decisions. The experimental design simulated an environment in which food was distributed in patches. Patches contained zero to four food items which could be collected by pecking at a key. All patches ended with sudden depletion. The time elapsed since the last reinforcement was the only way to detect the depletion of the patch. Once a patch was depleted, a new patch could be reached by completion of a travel requirement of 20 flights between two perches. Key pecks within a patch and the time of the last response in a patch (giving-in time) were recorded. The level of variability in the inter-reinforcement intervals was varied between different conditions. An increase in inter-reinforcement interval variability resulted in a flattening of response rate functions and giving-in time distributions, and in more asymmetry of the response functions, but not of the giving-in time distributions. Two theoretical models of decision making are presented, which differ in the assumptions about memory constraints. In one case, all inter-reinforcement intervals are remembered but in the other, only the intervals with extreme values are remembered. Both models accommodate response rates as a function of trial time, but only the second is compatible with the observed departure decision. Our results are compatible with net rate maximization.