The reconstruction of the corticospinal tract in the human brain is a clinically important task for both surgical planning and population studies. Diffusion MRI tractography provides an in-vivo and patientspecific technique for mapping the tract’s geometry; however, its relationship to other bundles, such as the superior longitudinal fasciculus, presents issues for the standard tensor model, as it cannot represent their crossing fibers. We explore multi-fiber models that have been shown to overcome some of these issues, and evaluate methods for improving on previous work with model-based filtering of orientations. We conduct experiments with real clinical data including normal and tumor-infiltrated corticospinal tracts and compare the single tensor, multi-tensor, and filtered multi-tensor approaches. We found the multi-fiber approach to allow for lateral projections of the tract to be reconstructed and found the addition of orientation filtering to reduce outlier fibers and increase the number of lateral projections. Our results suggest this approach could be considered for clinical applications of corticospinal tract modeling