Methodologic investigations have addressed selection and recall bias in case-control studies of diet and breast cancer, whereas the effect of disease progression and medical treatment on estimates of dietary intake has been largely overlooked. In a multicenter, population-based case-control study of breast cancer in the United States, 1,588 newly diagnosed cases and 1,451 controls completed a self-administered food-frequency questionnaire. Initial evaluation suggested increased risk related to high intakes of calories, carbohydrates, fat, and protein. All nutrient associations were diminished after adjustment for calories. Evaluation by stage of disease revealed no relation of calories to risk among women with in situ disease, but elevated risks among women with localized (odds ratio [OR] = 1.33, 95 percent confidence interval [CI] = 1.0-1.7 highest cflowest quartile) or regional and distant disease (OR = 1.79, CI = 1.3-2.4). Further evaluation showed that the increased risk associated with calories was restricted to cases who reported having been treated with chemotherapy (OR = 1.66, CI = 1.3-2.1). A gradient of increasing risk with time interval from diagnosis to interview suggested the chemotherapy regimen itself and not necessarily characteristics of tumors requiring this treatment was responsible for the observed increased risk. These results indicate that epidemiologic studies of diet and breast cancer, particularly among young women, should evaluate possible bias related to post-diagnosis influences.