In 1986 a prospective cohort study on diet and cancer was started in the Netherlands among 62,573 women ages 55-69 years. Baseline information on diet and other risk factors was collected with a questionnaire. Cancer incidence was measured by record linkage with cancer registries and a pathology register. A case-cohort approach was used, in which the accumulated person time in the cohort was estimated by follow-up of a randomly selected subcohort (n = 1812). After 3.3 years of follow-up, 471 incident breast cancer cases were available for analysis. Questionnaire data for these cases and the 1716 female subcohort members without a history of cancer other than skin cancer were analyzed. In a multivariate analysis, controlling for traditional risk factors, the relative rates for breast cancer in increasing quintiles of energy-adjusted total fat intake were 1.00, 1.00, 1.34, 1.22, 1.08 (P-trend, 0.32). For saturated fat there was some evidence for a weak positive association when quintiles were used (relative rates in quintiles 1-5, 1.00, 1.22, 1.22, 1.38, 1.39; P-trend, 0.049). The 95% confidence interval (CI) for the top quintile was 0.94-2.06, however; and when saturated fat was used as a continuous variable, the effect was no longer significant (P = 0.20). Relative rate estimates for the highest versus lowest quintiles of monounsaturated fat, polyunsaturated fat, and cholesterol intake were 0.75 (95% CI, 0.50-1.12), 0.95 (95% CI, 0.64-1.40) and 1.09 (95% CI, 0.74-1.61), respectively, with no evidence for significant trends. This prospective study does not support a major role of dietary fat in the etiology of postmenopausal breast cancer.