Abstract: The Marginal Value Theorem (MVT) describes the behaviour that maximizes the ratio of expected gain over expected foraging time in a patchy environment. When travel time is variable, the MVT rationale and its predictions are sensitive only to the mean travel time and not to the spread or skew of the distribution. Two mechanistic arguments contradict these predictions. First, tests of the MVT have previously shown that there is a disproportionate influence of the last travel time, and second, psychological models of information processing suggest that memory for time intervals is strongly dependent on the scatter of the distribution experienced. These mechanistic concepts, combined with Jensen's inequality, suggest that patch exploitation should decrease as the scatter of the travel distribution increases. In a Skinner box experiment with pigeons, Columba livia, the problem was examined by simulating three environments with identical patches and the same mean travel time, but different travel time variability. Patch exploitation decreased with increasing variance in travel time. The results are used to argue in favour of the inclusion of realistic psychological properties as constraints in functional models of behaviour. Although both the MVT and the mechanistic models account for some features of the results, none of them can explain all the findings.