Quantitative computed tomography (QCT) based nonlinear homogenized finite element (hFE) models of the human femur do not take bone׳s microstructure into account due to the low resolution of the QCT images. Models based on high-resolution peripheral quantitative computed tomography (HR-pQCT) are able to include trabecular orientation and allow the modeling of a cortical shell. Such a model showed improvements compared to QCT-based models when studying human vertebral bodies. The goal of this study was to compare the femoral strength prediction ability of subject specific nonlinear homogenized FE (hFE) models based on HR-pQCT and QCT images. Thirty-six pairs of femurs were scanned with QCT as well as HR-pQCT, and tested in one-legged stance (STANCE) and side-ways fall (SIDE) configurations up to failure. Non-linear hFE models were generated from HR-pQCT images (smooth meshes) and compared to recently published QCT based models (voxel meshes) as well as experiments with respect to ultimate force. HR-pQCT-based hFE models improved ultimate force (R(2)=0.87 vs 0.80, p=0.02) predictions only in STANCE configuration but not in SIDE (R(2)=0.86 vs 0.84, p=0.6). Damage locations were similar for both types of models. In conclusion, it was shown for the first time on a large femur dataset that a more accurate representation of trabecular orientation and cortex only improve FE predictions in STANCE configuration, where the main trabecular orientation is aligned with the load direction. In the clinically more relevant SIDE configuration, the improvements were not significant.