Using four Guinea pigs ( Cavia aperea f. porcellus) as subjects we tested the predictions of three versions of a rate-maximizing model in a patch choice foraging task. Patches were of two types which differed in interval between captures and in number of prey within each patch. In the model, the optimal policy is expressed as the threshold difference in inter-prey interval between the two patch types beyond which it pays to reject poor patches. The model versions are: a) recognizer (capable of identifying patch type on encounter and patch exhaustion upon capturing the last prey); b) timer (must wait a time r before identifying a patch as being poor, but can identify patch exhaustion at the last capture); and c) dual timer (as timer but needs an additional waiting time to detect patch exhaustion). In the experiment we used patches containing only one prey item. The results showed that patches were accepted in close agreement with the predictions of the timer and dual-timer models, but only the assumptions of the second model were met, as guinea pigs took considerable time to identify poor patches and also some time to abandon exhausted patches. These results showed that although cues for patch type were present, the guinea pigs did not behave as recognizers. We discuss and advocate the use of realistically constrained optimal foraging models.