We evaluate the accuracy of three well-known fracture growth theories to predict crack trajectories in anisotropic rocks through comparison with new experimental data. The results of 99 fracture toughness tests on the metamorphic Grimsel Granite under four different ratios of mixed-mode I/II loadings are reported. For each ratio, the influence of the anisotropy orientation on the direction of fracture growth is also analyzed. Our results show that for certain loading configurations, the anisotropy offsets the loading influence in determining the direction of crack growth, whereas in other configurations, these factors reinforce each other. To evaluate the accuracy of the fracture growth theories, we compare the experiment results for the kink angle and the effective fracture toughness with the predictions of the maximum tangential stress (MTS), the maximum energy release rate (MERR), and the maximum strain energy density (MSED) criteria. The criteria are first assessed in their classical forms employed in the literature. It is demonstrated that the energy-based criteria in their classical formulation cannot yield good predictions. We then present modified forms of the ERR and SED functions in which the tensile and shear components are decomposed. These modified forms give significantly better predictions of fracture growth paths. The evaluation of the three criteria illustrates that the modified MSED criterion is the least accurate model even in the modified form, while the results predicted by MTS and modified MERR are well matched with the experimental results.