In vivo bone densitometry is affected by measurement inaccuracies arising from the assumptions made about soft tissue and marrow composition. This study tested the hypothesis that section modulus (SM, a measure of bending resistance) when measured ex vivo, would discriminate cases of hip fracture from controls better than areal bone mineral density (aBMD). The biopsies were from (n = 22, female) subjects that had suffered an intracapsular hip fracture. The control material (n = 24, female) was from post-mortem subjects. Serial peripheral quantitative computed tomography (pQCT) 1-mm thick cross-sectional images of femoral neck previously embedded in methacrylate were obtained with the Densiscan 1000 pQCT densitometer and matched for lateral location. The image voxels were converted to units of bone mass, which were then used to derive the section modulus. The data were used to derive means from which receiver operating characteristic (ROC) curves could be generated. The area under the curves (AUC) showed that discrimination between the fracture cases and controls was better for SM than aBMD [SM: AUC = 0.83 (95% confidence interval: 0.71, 0.96), aBMD: AUC = 0.70 (0.54, 0.85); P = 0.034]. To simulate the forces experienced during a sideways fall, the model's neutral axis was rotated by 210 degrees. The results for section modulus were predictable from those at 0 degrees (r(2) = 0.97). We conclude that biomechanical analysis of the distribution of bone within the femoral neck may offer a marked improvement in the ability to discriminate patients with an increased risk of intracapsular fracture. Progress towards implementing this form of analysis in clinical densitometry should improve its diagnostic value, but may depend in part on better image resolution and more accurate corrections for the variability between subjects in regional soft tissue composition.