To determine whether estrogen initiated at age 60 yr or later reduces rates of bone loss and fracture incidence, we performed a prospective cohort study of 6910 nonosteoporotic women, 65 yr of age or older. Estrogen use, medical history, lifestyle, and anthropometric data were obtained by questionnaire, interview and examination. We identified five patterns of estrogen use: never users (67%); past early users (started under age 60 yr with no current use; 23%); past late users (started at age 60 or later with no current use; 2%); current early users (started under age 60 yr with use both at baseline and 6 yr later; 6.7%); and current late users (started at age 60 or later with use at baseline and 6 yr later; 1.5%). Bone mineral density was measured at the total hip twice, an average of 3.5 yr apart, and at the calcaneus, an average of 5.7 yr apart. Incident nonspine fractures were validated by radiographic report. Bone mineral density was significantly higher among current users, compared with never and past users. The annual rate of hip bone loss was significantly lower in current early users (-0.22%/yr) and current late users (-0.35%/yr) in comparison with never users (-0.6%/yr), past early users (- 0.6%/yr), and past late users (- 0.72%/yr). During an average of 11.0 yr of follow-up, 1953 nonspine fractures were confirmed. The multiple-adjusted relative risk of nonspine fracture was 0.63 (95% confidence interval 0.51-0.78) among current early users and 0.75 (0.50-1.12) among current late users, compared with never users. Early initiation and long-term continuation of estrogen is associated with a reduction in the risk of nonspine fractures, and initiation at or after age 60 yr with long-term continuation may also be associated with a reduced fracture risk.