Bone mineral density (BMD) has been shown to predict fracture risk in community-dwelling older persons; however, no comparable prospective study has been performed in the long-term care setting where the role of BMD testing is uncertain. To determine the ability of a single BMD measurement to predict the risk of subsequent fracture in long-term care residents, we designed a prospective study in a 725-bed long-term care facility. A total of 252 Caucasian nursing home residents (mean age 88 years, 74% women) were recruited between 1992 and 1998. BMD of the hip, radius or both sites was measured using dual-energy X-ray absorptiometry. Participants were followed through September 1999 for the occurrence of fracture. Cox proportional hazards regression models were constructed to determine the relationship between BMD and the risk of fracture controlling for potentially confounding variables. Sixty-three incident osteoporotic fractures occurred during a median follow-up time of 2.3 years. The multivariate-adjusted risk of fracture for each standard deviation decrease in BMD was 2.82 (95% CI 1.81-4.42) at the total hip, 2.79 (95% CI 1.69-4.61) at the femoral neck, 2.26 (95% CI 1.51-3.38) at the trochanter, 1.83 (95% CI 1.14-2.94) at the radial shaft and 1.84 (95% CI 1.21-2.80) at the ultradistal radius. Subjects in the lowest age-specific quartile of femoral neck BMD had over 4 times the incidence of fracture compared with those in the highest quartile. BMD at either hip or radius was a predictor of osteoporotic fracture, although in women, radial BMD did not predict fracture. Knowledge of BMD in long-term care residents provides important information on subsequent fracture risk.