Current methods for determining the risk of fracture for bones with metastatic lesions are inadequate. This research investigated the flexural and torsional behavior of femoral shafts with simulated lesions as a means toward improving clinical guidelines. Four-point bending tests demonstrated differences in the failure characteristics of whole bones and bones with hemispherical defects;​ whole bones exhibited greater structural ductility with five times the energy-to-failure (p < 0.01). Linear finite element (FE) models predicted failure loads for both sets of bones a priori and demonstrated the benefit of using computed tomography (CT) scan data to describe bone geometry and density-based heterogeneity. Models not utilizing CT scan data were less accurate and precise (r = 0.76) than models using CT scan data for geometry (r = 0.93) or for both geometry and heterogeneous material properties (r = 0.97). A parametric sensitivity analysis revealed that linear FE models could not explain the differences in structural behavior of whole bones and bones with defects. Nonlinear models that incorporated a bi-linear stress-strain relationship for cortical bone performed as well as linear models in predicting ultimate strengths for the flexural experiment (r = 0.99). Moreover, the behavior of these nonlinear models provided a possible explanation for differences in ductility between the two sets of bones;​ perhaps, the "brittle" failure of bones with defects was the consequence of early but concentrated plastic yielding, imperceptible on macroscopic load-deflection curves. CT scan-derived linear models were then used to predict the torsional strengths of bones with hemispherical defects. The linear FE models achieved high precision (r = 0.99), but overestimated ultimate torques by a factor of two. This was most likely the consequence of a failure criterion that did not account for the orthotropy of cortical bone. Future studies should address the issue of material property assumptions and failure criteria used in the modeling of cortical bone structures. Until that time, the precision of the CT scan-derived FE models of this investigation marks a significant step toward the prediction of failure for bones with metastatic lesions.