In recent years, it has been suggested that statistical shape model based 2D–3D reconstruction of the proximal femur could offer a solution to issues with preoperative planning of total hip replacement from 2D radiographs. The purpose of this work was to assess the effect of radiographic femoral rotation on the accuracy of a statistical shape model based 2D–3D femoral reconstruction method. A reconstruction algorithm was tested on input images with varying amounts of internal/external rotation (10 internal to 50 external) using leave-one-out tests and the resulting 3D shape was compared to the CT segmentation by point-to-point distance. The minimum value for mean point-to-point error was 1.24 ± 0.18 mm and occurred at 20º of external rotation (where neutral orientation was the femoral neck axis aligned in the coronal plane). The maximum error calculated was at 50º of external rotation with mean point-to-point error being 1.88 ± 0.41 mm. This work highlights an important source of error for 2D–3D reconstruction algorithms which may be incorporated into future validation studies.