We examined the association between menstrual cycle characteristics (cycle length, variability, and bleeding length) and physical and behavioral attributes in 766 women age 29-31 years. Menstrual cycle data were prospectively recorded as part of the Menstruation and Reproductive History Study of college women in Minnesota, begun by Alan Treloar in 1934. Data on lifetime height, weight, physical activity, alcohol and caffeine consumption, and smoking history were collected in 1990 using a self-administered questionnaire. Cycle variability, as measured by the standard deviation of the cycle length, was increased, and menstrual cycles > or = 42 days in length were more common among women in the lowest quartile of Quetelet index [odds ratio (OR) for long cycle = 1.6;95% confidence interval (CI) = 0.82-3.0] and among the most physically active (OR = 1.7;95% CI = 0.93-3.1). Long menstrual cycles were less common (OR = 0.40;95% CI = 0.22-0.73) among women who drank alcohol than among nondrinkers. Variable or long menstrual cycles may reflect anovulation and relatively low levels of estrogen exposure. We would expect, based on our data, reduced estrogen exposure among lean women, physically active women, and those who do not consume alcohol. These findings suggest an explanation for the reported associations between these factors and breast cancer risk.