Biological fixation of cementless femoral implants requires primary stability by optimal fit in the proximal femur. The anatomy of the bone must then be known precisely. We analysed in vitro the accuracy of bone measurements of 32 femurs and compared the dimensions obtained from radiographs and CT scans with the true anatomical dimensions. Standard radiographs gave only a rough approximation of femoral geometry (mean difference: 2.4 +/- 1.4 mm) insufficiently accurate to allow selection of the best fitting prosthesis from a range of sizes and altogether inadequate to design a custom-made prosthesis. CT scans give greater accuracy (mean difference: 0.8 +/- 0.7 mm) in our experimental conditions, but in clinical practice additional sources of error exist.