We describe a prospective study in which bone mineral density (BMD) was measured in total body and regions, proximal femur, lumbar spine, and forearm in 84 apparently normal postmenopausal women with normal spinal radiographs and in 47 women with 1-10 wedged or compressed vertebrae. There was a history of peripheral fracture in 35 of the 84 controls and 30 of the 47 osteoporotics (p < 0.02) but there was no association between vertebral fracture and wrist fracture. At all sites and regions, the differences in BMD between the "normal" and "osteoporotic" women was highly significant and all but "ribs" and "arms" remained significant after correction for menopausal age. In the whole set, and in both subgroups, the coefficients of correlation between sites and regions were all highly significant (p < 0.001). Nonetheless, some sites discriminated better between the two groups than others. Standardized odds ratios (OR) for vertebral fracture versus no-fracture were calculated by logistic regression and expressed as the rise in OR for each standard deviation (SD) fall in bone density. This ratio was greatest (3.4) in "pelvis" and weakest (1.7) in "ribs" but all were statistically significant. Geometric mean regression equations were calculated for all the 78 possible pairs of sites and regions in the 84 normal subjects and the deviations of the osteoporotic women from these normal slopes calculated. In most pairs of sites and regions, the vertebral fracture cases were scattered around the normal group's slope but fell lower down on both axes. The bone deficits in the osteoporotics compared with young normal women ranged from -14% in "head" to -40% in Ward's triangle and the T-scores ranged from -1.9 in "ribs" to -3.9 in the forearm. Sensitivity ranged from 17% in "ribs" to 36.2% in Ward's triangle. Specificity varied between 88 and 94% and the percent correctly classified ranged from 62.6% in "ribs" to 72.5% in Ward's triangle. We conclude that primary postmenopausal osteoporosis affects the entire skeleton but that some sites discriminate better between vertebral fracture and nonfracture cases regardless of whether they represent cortical or trabecular bone. (