The relation of oral contraceptive use to the risk of ovarian cancer was assessed with data collected during 1977-1991 from patients under 65 years of age in hospitals in Boston, New York, Philadelphia, and Baltimore. We compared 441 women with recently diagnosed invasive epithelial ovarian cancer to 2,065 control women. Logistic regression was used to control risk factors for ovarian cancer. The multivariate relative risk estimate decreased with the increasing duration of oral contraceptive use (p < 0.05): the estimate was close to 1.0 for duration categories of less than 3 years; it was reduced for the categories of 3-4 years of use and greater, but it did not decline further as the duration of use increased. For > or = 3 years of use, the estimate was 0.6 (95% confidence interval 0.4-0.8). The inverse association of risk with > or = 3 years of use was consistently present across categories of age, parity, interview year, and geographic area. It was apparent for as long as 15-19 years after cessation. Many different specific oral contraceptive formulations appeared related to a decreased risk; however, data were sparse for the newer types, particularly phasic preparations, and the ability to assess specific preparations in the context of use of multiple preparations was limited. The present data confirm previous reports of an inverse association of ovarian cancer risk with oral contraceptive use of several years in duration. They also suggest that the association may persist for as long as two decades and that it is not confined to any particular type of oral contraceptive formulation.