Femoral neck axis length, neck width, and neck-shaft angle were measured on radiographs of right proximal femora from 64 cadavers (28 female, 36 male). Bone mineral density (BMD) was measured using dual energy X-ray absorptiometry (DXA) for various regions of interest, and quantitative computed tomography (QCT) was used to determine BMD and bone areas for cortical and trabecular bone at the trochanter and femoral neck. The strength of the femur was determined by a mechanical test simulating a fall on the greater trochanter, and the fracture type (cervical or trochanteric) was subsequently determined from radiographs. Twenty-six cervical fractures and 38 trochanteric fractures were observed, with no significant sex difference in the distribution of fracture types. Femoral strength was significantly elevated in males compared to females. DXA trochanteric BMD was more strongly (p < 0.05) correlated with femoral strength (r2 = 0.88) than were any of the other DXA BMD measurements (r2 = 0.59-0.76). In multiple regression models, a combination of different DXA BMD measurements produced only a small increase (1%) in the explained variability of femoral strength. Of the QCT measurements, trochanteric cortical area yielded the optimal correlation with femoral strength (r2 = 0.83). Weak, but significant, correlations were observed between femoral strength and cortical BMD at trochanteric (r2 = 0.28) and neck regions (r2 = 0.07). In multiple regression models, combining QCT parameters yielded, at best, an r2 of 0.87. Of the geometrical parameters, both neck axis length and neck width were significantly correlated with femoral strength (r2 = 0.24, 0.22, respectively), but no significant correlation was found between strength and the neck-shaft angle. Combining DXA trochanteric BMD with femoral neck width resulted in only a small increase in the explained variability (1%) compared to trochanteric BMD alone. The results demonstrated that DXA and QCT had a similar ability to predict femoral strength in vitro. Trochanteric BMD was the best DXA parameter, and cortical area (not cortical BMD) was the optimal QCT parameter. Geometric measurements of the proximal femur were only weakly correlated with the mechanical strength, and combinations of DXA, QCT, and geometric parameters resulted in only small increases in predictive power compared to the use of a single explanatory variable alone.