SUMMARY: Incident hip fractures and non-hip osteoporotic fractures were studied in 30,953 women during mean 3.7 years of observation. Hip axis length (HAL) and strength index (SI) made a small but statistically significant contribution to hip fracture prediction that was independent of age and hip bone density. INTRODUCTION: It is uncertain whether bone geometric measures improve fracture prediction independent of conventional areal bone mineral density (BMD). METHODS: Women aged > or =50 years with hip dual-energy x-ray absorptiometry were identified from the regionally based database in the Province of Manitoba, Canada. Scans were reprocessed to derive parameters of hip bone geometry. Incident hip fractures (N = 270) and non-hip osteoporotic fractures (N = 1,347) were identified during mean 3.7 years of observation. RESULTS: HAL was greater in both hip and non-hip fracture cases than in non-fracture cases, whereas cross-sectional moment of inertia, cross-sectional area, and femoral SI were all significantly less. After adjustment for total hip BMD, HAL [hazard ratio (HR) 1.22 per SD increase, 95% CI 1.07-1.38] and SI (HR 1.21 per SD decrease, 95% CI 1.07-1.37) were independent predictors of hip fractures but not of non-hip fractures. When both HAL and SI were added to a model containing age and total hip BMD, there was a small improvement in hip fracture prediction (ROC area under the curve 0.832 +/- 0.013 vs 0.823 +/- 0.013; P = 0.001). CONCLUSIONS: HAL and SI made a small but statistically significant contribution to hip fracture prediction that was independent of age and BMD measurement.