The patch choice model has four main assumptions (Kelly 1995). First, in contrast to the diet breadth model, the patch choice model assumes that resources are distributed heterogeneously, or in patches across the landscape. Second, these patches will be encountered at random, in proportion to their frequency. Third, a forager will not return to a previously foraged patch until the resources within that patch are rejuvenated. Fourth, the time a forager spends traveling from one patch to another is unproductive (that is, no potential food resources are encountered between patches). The main point of inquiry with the diet breadth model is: at what point should an optimal forager stop foraging for resources in one patch and move to forage in another patch?
To address the fact that, in most environments, return rates will progressively diminish as a forager continues to collect resources from within a patch, Charnov (1976) derived the Marginal Value Theorem. This theorem predicts that “the optimal forager should leave any patch when it is depleted to the point where foraging elsewhere will yield higher returns per unit time (taking travel costs into account)”(Smith 1983:631).
Again, there are numerous weaknesses associated with both the patch choice model and the marginal value theorem. Due to the assumptions of the patch choice model, there are no true anthropological tests of either the model or the theorem. Primarily, foragers do not encounter patches at random; rather, they will decide where to forage before setting out. Additionally, the assumption that time spent traveling between patches is unproductive is flawed. Rarely is it the case that a forager will not encounter either a potential food resource or evidence of a resource (i.e. tracks, etc) while moving between patches.
Similar to linear programming, the patch choice model requires that the researcher have extensive knowledge about the environment (i.e. distance between patches, density of possible food resources within the patches, etc.) and is therefore difficult to apply archaeologically. Again, as with linear programming, Sheehan expresses the opportunity paleoethnobotanical research offers in reconstructing prehistoric environments.