Vertebral fractures are common in the elderly, but efforts to reduce their incidence have been hampered by incomplete understanding of the failure processes that are involved. This study’s goal was to elucidate failure processes in the lumbar vertebra and to assess the accuracy of quantitative computed tomography (QCT)-based finite element (FE) simulations of these processes. Following QCT scanning, spine segments (n = 27) consisting of L1 with adjacent intervertebral disks and neighboring endplates of T12 and L2 were compressed axially in a stepwise manner. A microcomputed tomography scan was performed at each loading step. The resulting time-lapse series of images was analyzed using digital volume correlation (DVC) to quantify deformations throughout the vertebral body. While some diversity among vertebrae was observed on how these deformations progressed, common features were large strains that developed progressively in the superior third and, concomitantly, in the midtransverse plane, in a manner that was associated with spatial variations in microstructural parameters such as connectivity density. Results of FE simulations corresponded qualitatively to the measured failure patterns when boundary conditions were derived from DVC displacements at the endplate. However, quantitative correspondence was often poor, particularly when boundary conditions were simplified to uniform compressive loading. These findings suggest that variations in trabecular microstructure are one cause of the differences in failure patterns among vertebrae and that both the lack of incorporation of these variations into QCT-based FE models and the oversimplification of boundary conditions limit the accuracy of these models in simulating vertebral failure.