The suitability of depletion theory for predicting the distribution and movements of wintering brent geese feeding in habitat patches containing foods differing qualitatively as well as quantitatively is evaluated. By monitoring both digestibility and nutrient content of potential foods throughout the season, we assess profitability of habitat patches using assimilation rates. We argue that these geese do not conform to the predictions of an ideal free distribution because they are constrained both by nitrogen limitation and perceived mortality risks. Instead, for most of the season they exhibited partial feeding preferences by feeding on two or more types of food each day. They fed on salt marsh plants throughout the entire wintering season. In addition, from October until March they fed for part of each day on supplementary sites that were more profitable for nitrogen. In October they fed first on intertidal algae, the most profitable source of nitrogen. When this became depleted in late autumn, they moved inland to feed initially on winter wheat, where they were subject to control shooting, then onto pastures. By mid-March the pastures were no longer a significantly more profitable source of nitrogen. The geese then switched to feeding only on the salt marshes at a cost of a 39% decrease in their overall assimilation rates. The nitrogen limitation hypothesis was supported by results of experimentally altering the nitrogen content of pasture swards. Feeding preferences correlated positively with changes in nitrogen content, but not water-soluble carbohydrate content of experimental swards. We conclude that predictions of simple depletion models are unlikely to explain the movements of herbivores between patches that differ in digestibility and nutrient content as well as in the quantity of foods available and that multi-currency models are a more appropriate means of predicting foraging behaviour of herbivores exhibiting partial feeding preferences.