Learning animals are predicted to use a flexible patch-leaving threshold (PLT) while foraging in a depletable environment under exploitative competition. This prediction was tested in flock-feeding Bewick's swans (Cygnus columbianus bewickii) depleting hidden tubers of fennel pondweed (Potamogeton pectinatus) in a two-dimensional, continuous environment. The swans' patch residence time was measured by combining recordings of the foraging behaviour and movement paths. The tuber biomass density was measured before and after the period of exploitation, using the presumable foraging window of the swans as the scale of measurement. Swan foraging was simulated in order to predict the effects of flexible and fixed PLTs, respectively, on the patch residence time and the spatial heterogeneity of the tuber biomass density. Flexible PLTs were predicted to lead to short and decreasing patch residence times and a decrease in the coefficient of variation in tuber biomass densities, whereas the reverse was generally the case for fixed PLTs. Observed patch residence times did not decrease with time and were intermediate between those predicted for swans with flexible and fixed PLTs. Furthermore, an increase of the coefficient of variation in the tuber biomass density was observed. Given the observed giving-up biomass densities the most likely model was one with swans with a fixed rather than a flexible PLT. These results point at factors that may affect the spacing behaviour or constrain the use of a flexible PLT in swans.