Structural tests, such as whole bone torsion tests, have become widely accepted methods for assessing average bone material properties. To simplify interpretation of these tests, the nonuniform bone geometry is often analyzed as a tube with a constant cross section (prismatic) and the areal properties of the smallest bone section. This approach may not adequately represent the true torsional behavior of the cross section and does not account for any lengthwise variations in bone geometry. The errors introduced by these approximations are particularly significant when comparing bones of different sizes and geometries. In this paper, we examine the effects of approximating the cross-sectional torsional behavior and of neglecting lengthwise variations in bone geometry. We then present a simple, standardized procedure utilizing a FORTRAN computer program for accurate determination of material properties. We examine first simple idealized bone geometries and then a complex three-dimensional model of the femur from a 26-day-old male Sprague-Dawley rat. For these models, the conventional methods for interpreting torsion tests introduce errors of up to 42% in the shear modulus and up to 48% in the maximum shear stress; a straightforward extension of these methods reduces the errors to within 3%.