Menstrual cycle characteristics may reflect underlying endocrine patterns that influence the risk of breast cancer. Most previous studies of menstrual function and breast cancer risk have used retrospective reports of menstrual bleeding, which may be unreliable. To examine this association, the authors conducted a mail survey among 997 women who had recorded menstrual events prospectively over as many as 50 years, beginning in 1934. Compared with women with a median menstrual cycle length of 26-29 days, women who had cycles of extreme length at ages 25-29 years had a nearly twofold increased incidence of breast cancer (for a median cycle length of less than 26 days, adjusted relative risk (RR) = 1.9, 95% CI 0.9-4.1; for > or = 34 days, RR = 1.9, 95% CI 0.9-3.9). Statistical adjustment was made for age, family history of breast cancer, parity, age at menopause, age at first pregnancy, and Quetelet index (weight (kg)/height (m)2). Adjusting for age and other potential confounders and restricting the analysis to women who did not use hormones, women who experienced either a lesser (< 150) or a greater (> 350) cumulative number of cycles had an increased incidence of breast cancer (adjusted RR = 1.9, 95% CI 0.3-10.6, and RR = 1.8, 95% CI 0.5-6.0, respectively) compared with women who experienced 150-350 cycles. The findings are discussed in the context of current hormonal theories of breast cancer etiology.