Three-dimensional accurate evaluation of the geometry of the proximal femur may be helpful for hip fracture risk evaluation. The purpose of this study was to apply and validate a stereo-radiographic 3D reconstruction method of the proximal femur, using contours identification from biplanar DXA images. Twenty-five excised human proximal femurs were investigated using a standard DXA unit. Three-dimensional personalized models were reconstructed using a dedicated non-stereo corresponding contours (NSCC) algorithm. Three-dimensional CT-scan reconstructions obtained on a clinical CT-scan unit were defined as geometric references for the comparison protocol, in order to assess accuracy and reproducibility of the 3D stereo-radiographic reconstructions. The precision of a set of 3D geometric parameters (femoral-neck axis length, mid-neck cross-section area, neck-shaft angle), obtained from stereo-radiographic models was also evaluated. This study shows that the NSCC method may be applied to obtain 3D reconstruction from biplanar DXA acquisitions. Applied to the proximal femur, this method showed good accuracy as compared with high-resolution personalized CT-scan models (mean error = 0.8 mm). Moreover, precision study for the set of 3D parameters yielded coefficients of variation lower than 5%. This is the first study providing 3D geometric parameters from standard 2D DXA images using the NSCC method. It has good accuracy and reproducibility in the present study on cadaveric femurs. In vivo prospective studies are needed to evaluate its discriminating potential on hip fracture risk prediction.