In the spine, cancellous bone quality is usually assessed for the whole vertebral body in a transverse central slice. Correct identification and assessment of the weakest parts of the cancellous bone may lead to better prediction of fracture risk. The density and structural parameters were systematically investigated inside the thoracic (T-9), thoracolumbar (T12–L1), and lumbar (L-4) vertebral bodies of nine subjects. On both sides of the median sagittal plane, anterior and posterior 8.2 mm vertical cores were harvested in the thoracic vertebra. In the thoracolumbar and lumbar vertebrae, external samples were also cored. Peripheral quantitative computed tomographic (pQCT) density analysis of the 136 cores was performed at four different levels, from the lower to the upper endplate. The relatively thin slice thickness (300 μm) and small pixel size (70 μm × 70 μm) was considered sufficient to investigate the structural parameters on the four transverse slices and in the sagittal and coronal planes (total of 816 images). Using a constant threshold a binary image was generated and the morphometric data were extracted. The binary image was further skeletonized and classical strut analysis was performed. Cancellous bone density was 20% higher in the posterior cores than in the anterior and external cores. Moreover, clear vertical inhomogeneity was noted because the lowest half of the vertebral body presented lower density than the upper half (differences ranging from 25% to 15%). All structural parameters were strongly dependent on the location of the measurement. Structural differences between anterior, posterior, and external areas were mild and followed the density patterns. On the other hand, vertical inhomogeneity of the structural parameters was important. For example, in the thoracolumbar and lumbar vertebrae, the numbers of nodes or node-to-node struts were almost twofold higher in the inferior half than in the superior half (p < 0.01), whereas trabecular thickness and number of free-ends presented a center/close-to-endplate structural pattern, with central trabeculae being 15% thicker (p < 0.05) and presenting 30% fewer free-ends (p < 0.01) than the close-to-endplate ones. Variability of density and structural parameters was high and a substantial part of this variability could be explained by the place inside the vertebral body where the measurement was made. The weak part was not in the center of the body but in its upper half where the lower density did not seem to be compensated by a higher structural architecture. Further clinical investigation could enhance fracture prediction by tracking and focusing on the weakest part of the vertebral body.