The purposes of this study were to examine fracture risk profiles at specific bone sites, and to understand why model discrimination using clinical risk factors is generally better in hip fracture models than in models that combine hip with other bones. Using 3-year data from the GLOW study (54,229 women with more than 4400 total fractures), we present Cox regression model results for 10 individual fracture sites, for both any and first-time fracture, among women aged >/=55 years. Advanced age is the strongest risk factor in hip (hazard ratio [HR] = 2.3 per 10-year increase), pelvis (HR = 1.8), upper leg (HR = 1.8), and clavicle (HR = 1.7) models. Age has a weaker association with wrist (HR = 1.1), rib (HR = 1.2), lower leg (not statistically significant), and ankle (HR = 0.81) fractures. Greater weight is associated with reduced risk for hip, pelvis, spine, and wrist, but higher risk for first lower leg and ankle fractures. Prior fracture of the same bone, although significant in nine of 10 models, is most strongly associated with spine (HR = 6.6) and rib (HR = 4.8) fractures. Past falls are important in all but spine models. Model c indices are >/=0.71 for hip, pelvis, upper leg, spine, clavicle, and rib, but