This dissertation addresses four specific issues: modeling of the vertebral endplate, mesh generation for human vertebrae, material property variation in vertebral trabecular bone, and modeling spinal fusions in rabbits. The objective of this study was to develop and validate techniques for subject-specific finite element modeling and analysis of spinal constructs. While models have been applied to analyze whole bones and bone-implant constructs, application to fused spinal segments has been limited. Using computed tomography scans as input, models could provide a non-invasive means to assess biomechanics of fusion progress. Geometric segmentation and meshing techniques were developed for human vertebral endplate surface. Smooth B-spline surface representation was generated for the vertebral endplate surface. Smoothness was found to be an important guideline for judgment of a sound surface fit. A moderate decrease of resolution (close to the level of clinical scans) had minimal effect on errors. A set of atlas based techniques were developed to segment geometry and construct hexahedral finite element meshes of human vertebrae from CT scans. Algorithms were robust to low quality medical images, and allowed manipulation of the finite element meshes. Application of an atlas will allow automatic processing of a large number of subjects, and facilitate comparison between subjects within a study. Material properties are an essential input for the finite element models. The study investigated the effect of heterogeneity in the vertebral trabecular bone on the mechanical properties predictions for accurate material modeling in the finite element models of spine. The relationship between trabecular bone volume fraction and elastic constants was assessed. The results showed that inclusion of morphology can improve the predictive strength compared to a single power-law, and most of the variations in the power-law prediction can be captured by incorporating degree of anisotropy (DA) in the model. The finding may be used to establish intrasite relationships of density-modulus within the vertebral body to improve the material modeling in the finite element analysis of spinal constructs. The modeling techniques were implemented and validated by rabbit spinal fusion study. Posterolateral lumbar spinal fusion was performed on twelve animals. Fused segments harvested from two postsurgical time stages were mechanically tested. The tensile tests were simulated using finite elements with density based material modeling to represent the heterogeneity in the fusion mass. The model prediction of stiffness, which is the primary variable of interest in spinal fusion, strongly correlated with the measured data. The results demonstrated the feasibility of using the finite element method to estimate the biomechanics of spinal fusion. Although strength wasn't correlated, application of alternate material models showed the potential to improve strength predictions by choosing appropriate density and strength calibrations. Overall, this study showed that subject-specific finite element model has great potential for the study of spinal fusion mechanics.