Bone mineral density (BMD) and bone microarchitecture are important determinants for the mechanical properties of cancellous bone. Although BMD alone is a good predictor of average mechanical properties of cancellous bone, there remains unexplained variation in mechanical properties that might be due to missing information regarding bone microarchitecture. Recent developments in three-dimensional (3D) structural analysis have provided possibilities for measuring a variety of structural indices to characterize bone microarchitecture. The objectives of this study were to calculate structural indices and elastic constants of human cancellous bone from different skeletal sites and to investigate the predictive value of different 3D structural indices for the elastic properties of bone. A total of 237 cancellous bone samples taken from the iliac crest, lumbar spine, femoral head, and calcaneus were imaged with a 3D microcomputed tomography (μCT) system. The segmented 3D images were used to calculate BV/TV, BS/TV, Tb.Th, Tb.Sp, Tb.N, and MIL ratio and for microstructural finite-element (μFE) analysis to calculate Young’s moduli, shear moduli, and Poisson’s ratios. A subgroup of “critical” specimens within each site was selected to represent specimens that could not be identified as osteoporotic or normal on the basis of BMD measurement alone. For these “critical” specimens, structural indices and elastic constants were correlated by means of linear multivariate regression analysis. It was found that the elastic constants clearly correlated better when one of the 3D structural indices was included as independent variable than when BV/TV was the only independent variable. Each of the examined structural indices could improve the correlation:​ the R2 values were maximally increased from 53% (BV/TV alone) to 82% (BV/TV and MIL ratio). The most effective indices, however, were not the same for the different skeletal sites. Even better correlations were found when more than one of the 3D structural indices were included as independent variables:​ the R2 values were maximally increased from 53% (BV/TV alone) to 92% (BV/TV, Tb.Sp, and MIL ratio). The prediction of elastic constants for cancellous bone samples is clearly improved when BV/TV is supplemented with 3D structural indices. These results suggest that the determination of mechanical properties of bone and the diagnosis of osteoporosis can be improved if, in addition to BMD, the 3D bone microarchitecture is assessed in vivo.