In order to evaluate validation and application of energy gain functions in mammalian herbivores, the fresh clover (Trifolium repens) leaf was used as the food preferred by voles (Microtus fortis), and clover leaf patches were devised by maintaining leaf biomass constant and varying the leaf size for monitoring the vole′s foraging behavior. Sigmoid gain function was not detected in any patches. However, linear function models accurately predicted the vole′s residence time in the small and medium food patches. Although piecewise linear and asymptotic models fitted to observed food intakes by the voles foraging in the large food patches, the former precisely predicted residence time of the voles foraging in these patches. The linear and piecewise linear models were all speculated to be developed on the basis of the bite model, a functional response model representing a mechanism for regulating intake rates, which was considered to be a mechanistic model. Therefore, the primary purpose of the linear and piecewise linear models was to interpret the relationship between energy gain and residence time. In contrast, the asymptotic and sigmoid models were primarily empirical, representing no mechanisms at all. They predicted energy gain and residence time in the patch scales, and did not interpret the dynamics of intake rates, so they failed to predict accurately. On the other hand, these 4 models all included non-foraging activity time such as vigilance and escape time, which broadly limited their applications. Therefore, novel models for predicting energy gain and residence time are urgently needed in foraging ecology of mammalian herbivores.