In order to determine which clinical, anthropometric, dietary, and biochemical variables are independent predictors of total and regional bone mineral density (BMD) in normal postmenopausal women, a cross-sectional study of 140 normal postmenopausal women has been carried out. Subjects were white, aged 45-71 yr (mean 58 yr), and had no history of disorders or medication use likely to influence bone or calcium metabolism. Multiple regression analysis was used to derive models for total and regional BMD in terms of the other variables measured. The analysis indicated that total body BMD was positively related to fat mass (P less than 0.0001), serum estrone (P = 0.0095) and age at menopause (P = 0.0165), and negatively related to age (P less than 0.0001), 24-h urine calcium (P = 0.0002), sex hormone-binding globulin (P = 0.0003), and serum alkaline phosphatase activity (P = 0.0029) (R2 = 0.61). Similar relationships were found in the subregions of the total body scans and in the lumbar spine and proximal femur, with insulin-like growth factor-1, parity, and age at menarche also being related to BMD at at least two of these sites. Lean body mass was not an independent correlate of BMD at any site once fat mass was taken into account. Muscle strength, physical activity, alcohol intake, and dietary intakes of calcium, sodium and protein did not emerge as significant predictors of BMD in this homogeneous group of postmenopausal women. We conclude that total body fat is the most significant predictor of BMD throughout the skeleton and this relationship is not explicable in terms of either estrone production in fat tissue or the dependence of skeletal load-bearing on fat mass. The mechanism underlying this relationship is an important question to be addressed in bone biology.