With stepwise multiple logistic regression (MLR), probabilistic classification equations were developed to identify asymptomatic women who are at risk for development of fracture of the spine. Clinically normal women with low TBCa/square root H ratios can be classified as at risk for osteoporosis prior to their developing spinal compression fractures. With receiver operating characteristic (ROC) analysis, it was possible to verify the accuracy of the MLR model to discriminate "normal" women at risk, with high sensitivity and specificity. With the MLR model, discrimination of osteoporotic women (50-59 years) was made correctly for 86.2% of the total osteoporotic subjects with the TBCa data. Similar models were derived from the photon absorptiometry data. From the spinal density (BDs) data, correct classification in the 50-59 year group was 55.6% of the total osteoporosis subjects; from the radius density (BMCr) data, the corresponding value was 31%. The highest probability of identifying osteoporosis in all age categories was, therefore, on the basis of TBCa data. Similar, but less accurate discrimination was achieved with the BDs and BMCr data. These conclusions were confirmed by the application of receiver operating characteristic (ROC) analysis. Correct identification of the population at risk permits the timely and efficient application of therapeutic programs prior to onset of fracture. In a serial study of 104 peri-menopausal women, for example, it was possible to determine the P value for individuals measured annually over a 3-10 year period and thus to predict normal individuals at risk for developing osteoporosis each year.