Samples such as regurgitated pellets and food remains have traditionally been used in studies of bird diets, but these can produce biased estimates depending on the digestibility of different foods. Stable isotope analysis has been developed as a method for assessing bird diets that is not biased by digestibility. These two methods may provide complementary or conflicting information on diets of birds, but are rarely compared directly. We analyzed carbon and nitrogen stable isotope ratios of feathers of Glaucous Gull (Larus hyperboreus) chicks from eight breeding colonies in northern Alaska, and used a Bayesian mixing model to generate a probability distribution for the contribution of each food group to diets. We compared these model results with probability distributions from conventional diet samples (pellets and food remains) from the same colonies and time periods. Relative to the stable isotope estimates, conventional analysis often overestimated the contributions of birds and small mammals to gull diets and often underestimated the contributions of fish and zooplankton. Both methods gave similar estimates for the contributions of scavenged caribou, miscellaneous marine foods, and garbage to diets. Pellets and food remains therefore may be useful for assessing the importance of garbage relative to certain other foods in diets of gulls and similar birds, but are clearly inappropriate for estimating the potential impact of gulls on birds, small mammals, or fish. However, conventional samples provide more species-level information than stable isotope analysis, so a combined approach would be most useful for diet analysis and assessing a predator's impact on particular prey groups.