Abstract The purpose of this work was to determine the capacity of peripheral quantitative computed tomography (pQCT) to accurately measure the true physical properties of vertebral cancellous bone samples and to predict their stiffness. pQCT bone mineral density (BMD) was first measured in “ideal” conditions. Ten cubic specimens of vertebral cancellous bone (10 × 10 × 10 mm) were washed with a water jet, defatted, and scanned in saline after elimination of air bubbles; thirteen slices were obtained. Seventy-one unprepared cylindrical samples were scanned in more “realistic” conditions, which allow further biomechanical testing. After extraction from the vertebral body, the samples were pushed into a plastic tube (no effort was made to remove the marrow or air bubbles), and only four slices were obtained to reduce the duration of scan. For the 81 samples, the true bone volume fraction ( BV/TV, %), true apparent density (ρ app, g/cm 3), and tissue density (ρ tiss, g/cm 3) (an indicator of the degree of mineralization of the matrix) were then measured using Archimedes' principle. ρ app was closely correlated to BV/TV ( r 2 = 0.97). ρ tiss (1.58 ± 0.08 g/cm 2) was almost constant but had some influence on ρ app ( r 2 = 0.03, p < 0.001). The pQCT BMD predicted accurately ρ app ( r 2 = 0.96) and BV/TV ( r 2 = 0.93) for the cylinders. For the cubes, in ideal conditions, the same correlations were even better ( r 2 > 0.99, both). Analysis of covariance indicated no difference ( p > 0.05) in the regressions due to preparation of the samples. The stiffness was better predicted by the true ρ app ( r 2 = 0.87) than by BV/TV ( r 2 = 0.83), indicating that stiffness was influenced by small differences in the tissue density. Consequently, the correlation between pQCT BMD and stiffness was excellent ( r 2 = 0.84). The fact that pQCT did not ignore this “tissue density information” compensated for the inaccuracies linked to “realistic” scanning conditions of the cylinder.