Decades of research have produced substantial data on elk (Cervus elaphus) diets in winter, when foraging conditions are most likely to affect population dynamics. Using data from 72 studies conducted in western North America between 1938 and 2002, we collated data on elk diets and environmental variables. We used these data to quantify diet selection by elk and to test whether variation in elk diets is associated with habitat type, winter severity, period of winter, human hunting, and study method. Graminoids (grasses and grass-like plants such as sedges) dominated elk diets and consistently occurred at a higher proportion in the diet than in elk foraging habitats, indicating preference. Forbs commonly made up ≤5% of the diet, with no evidence for preference; we conclude that forb use is largely incidental to grazing for graminoids. Browse was consumed in proportion to its availability, implying that the amount of browse in the diet was primarily determined by habitat use rather than selection. Comparing the diets of elk and sympatric ruminants, elk consistently selected graminoids more strongly than sympatric ruminants with the exception of bison (Bison bison), suggesting that elk are not environmentally forced to adopt the graminoid-biased diet that they normally select. The proportion of open meadows and grasslands on winter ranges was strongly and positively associated with graminoid consumption by elk. The proportion of graminoids in the diet was significantly lower in elk experiencing severe winter conditions or predation risk from human hunting. The period of winter (early, middle, and late) had only small effects on elk diets, as did the method by which the diet was determined. Overall, variation in elk diets is well-explained by a consistent tendency to select graminoids if available, modified by winter habitat type, predation risk, and winter severity, which can constrain habitat selection and access to grazing opportunities. To fully understand variation in foraging behavior, biologists should recognize these broad patterns when interpreting resource selection data. Managers should recognize that inconspicuous behavioral responses to environmental stimuli can alter the diet in ways that probably carry nutritional consequences.