Recall and selection bias are well-recognized potential problems in case-control studies of alcohol and cancer, but few analyses have attempted to assess the direction and the magnitude of these potential biases. We thus examined alcohol consumption in relation to risk of breast cancer using dietary questionnaires administered both before and after the diagnosis of breast cancer in the Nurses' Health Study (United States). Among cohort members who completed a dietary questionnaire in 1986 and who were free of cancer, 616 were diagnosed with breast cancer during follow-up to December 1989. These cases and 1,277 controls (a random sample of cohort members who did not develop cancer up to 1990) then were sent another questionnaire inquiring about their diet in 1985. Four hundred and ninety-four cases (80.2 percent) and 999 controls (78.2 percent) responded to the second questionnaire. The analysis based on the prospective (1986) questionnaire demonstrated an elevated risk of breast cancer among women who drank 30 or more g of alcohol daily (about two drinks) relative to nondrinkers (odds ratio [OR] = 1.55, 95 percent confidence interval [CI] = 1.01-2.39). The analysis based on the retrospective questionnaire also indicated a similar but slightly attenuated elevation of risk of breast cancer among women who drank at least 30 g daily (OR = 1.42, CI = 0.85-2.40). In these data, bias due to selection and recall had only minor effects on reported intake of alcohol consumption.