A population-based case-control study of breast cancer with a focus on premenopausal women under 45 years of age, conducted in three geographic regions of the United States, enabled the evaluation of risk in relation to varying breastfeeding practices. Among premenopausal parous women (1,211 cases, 1,120 random-digit-dialing controls), a history of breastfeeding for two or more weeks was associated with a relative risk (RR) of 0.87 (95 percent confidence interval [CI] = 0.7-1.0). This relationship was not altered substantially by removing from the reference group women who had problems with breastfeeding in the first two weeks, including those with insufficient milk production. Risk was not related substantially to number of children breastfed or length of breastfeeding, although a relatively low risk was observed among those breastfeeding for the longest duration examined (RR = 0.67, CI = 0.4-1.1 for an average period per child of 72 or more weeks). Women who began to breastfeed at a young age (< 22 years) experienced the greatest reduction in risk, but other timing parameters (e.g., interval since first or last breastfeeding) were not predictive of risk. Risks were not modified substantially by age or menopause status, although the number of menopausal subjects examined was limited. Use of medications to stop breast milk was unrelated to risk (RR = 1.04). The results of this study do not support the notion that breastfeeding substantially reduces breast cancer risk; however, this may reflect the fact that most of our study subjects breastfed only for limited periods of time (average breastfeeding per child of 30 weeks). Further studies are needed to clarify the relationship of breastfeeding to breast cancer risk, and to determine possible etiologic mechanisms underlying any observed associations.