The objective of this study was to report our quantitative computed tomography (QCT) density-mechanical property regressions for trabecular bone for use in biomechanical modelling of the human spine. Cylindrical specimens of human vertebral trabecular bone (from T10 to L4) were cored from 32 cadavers (mean ± SD age = 70.1 ± 16.8; 13 females, 19 males) and scanned using QCT. Mechanical tests were conducted using a protocol that minimized end-artifacts over the apparent density range tested (0.09–0.38 g/ cm3). To account for the presence of multiple specimens per donor in this data set, donor was treated as a random effect in the regression model. Mean modulus (319 ± 189 MPa) was higher and mean yield strain (0.78 ± 0.06%) was lower than typical values reported previously due to minimization of the end-artifact errors. QCT density showed a strong positive correlation with modulus (n = 76) and yield stress (r2 = 0.90–0.95, n = 53, p < 0.001). There was a weak positive linear correlation with yield strain (r2 = 0.58, n = 53, p = 0.07). Prediction errors, incurred when estimating modulus or strength for specimens from a new donor, were 30–36% of the mean values of these properties. Direct QCT density-mechanical property regressions gave more precise predictions of mechanical properties than if physically measured wet apparent density was used as an intermediate variable to predict mechanical properties from QCT density. Use of these QCT density-mechanical property regressions should improve the fidelity of QCT-based biomechanical models of the human spine for whole bone and bone-implant analyses. © 2002 Orthopaedic Research Society. Published by Elsevier Science Ltd. All rights reserved.