Currently, few elderly women have a measurement of bone mineral density (BMD). The aim of this study was to assess the potential value of a two-step screening process to identify the elderly women who are at greatest risk of fracture because of very low BMD: (1) use simple clinical criteria to select women who are highly likely to have a very low BMD and (2) measure the BMD of the women so selected. We used baseline data from 6958 women aged 75 years or older who were participants in the EPIDOS prospective study of risk factors for hip fracture. The outcome variable was very low BMD measured at the femoral neck by dual-energy X-ray absorptiometry and defined as a T-score < or = -3.5. The factors most predictive of very low BMD were low weight, history of fracture after the age of 50 years, slow gait, balance impairment, low grip strength, and dependence for instrumental activities of daily living. A score based on the risk function including these factors has a sensitivity of 80% at the median cut-off. Hence, by measuring the BMD of only half the population, 80% of the women with very low BMD can be identified. Weight is the strongest determinant of very low BMD and has approximately the same sensitivity as the complete score. In conclusion, a risk score for very low BMD based on simple criteria such as weight could be a useful clinical tool to select elderly women for bone densitometry.