Depletion of experimental seed patches by granivorous animals often is used as a qualitative assay of foraging activity. An optimal foraging model suggests that seed amounts remaining when foragers leave patches (“giving-up-density”, GUD) also provide quantitative measures of foraging economics, diet strategies and foraging abilities. Such quantitative uses of GUDs rest on several largely untested assumptions. We tested two of these with Merriam’s kangaroo rats: that gain curves are smoothly decelerating, and that foragers leave patches at a constant harvest rate. Harvest rates indeed declined with patch residence time, but in the piecewise linear fashion expected of systematic search. Animals also revisited areas within patches less frequently than expected with random search. In the field, they depleted patches in multiple visits and did not use a constant-rate leaving rule. These deviations from model assumptions cast doubt on inferences about foraging ecology that have been based on quantitative GUD theory.