The correlation between bone mineral density and vertebral strength is not based on mechanical principles and thus the method cannot reflect the effects of subtle geometric features and densitometric inhomogeneities that may substantially affect vertebral strength. Finite element models derived from quantitative computed tomography (QCT) scans overcome such limitations. The overall goal of this study was to establish that QCT-based "voxel" finite element models are better predictors of vertebral compressive strength than QCT measures of bone mineral density with or without measures of cross-sectional area. QCT scans were taken of 13 vertebral bodies excised from 13 cadavers (L1-L4; age: 37-87 years; M = 6, F = 7) and used to calculate bone mineral density (BMD(QCT)). The QCT voxel data were converted into linearly elastic finite element models of each vertebra, from which measures of vertebral stiffness and strength were computed. The vertebrae were biomechanically tested in compression to measure strength. Vertebral strength was positively correlated with the finite element measures of strength (r(2) = 0.86, P < 0.0001) and stiffness (r(2) = 0.82, P < 0.0001), the product of BMD(QCT) and vertebral minimum cross-sectional area (r(2) = 0.65, P = 0.0008), and BMD(QCT) alone (r(2) = 0.53, P = 0.005). These results demonstrate that highly automated "voxel" finite element models are superior to correlation-based QCT methods in predicting vertebral compressive strength and therefore offer great promise for improvement of clinical fracture risk assessment.