Four different spinal cord injury (SCI) models (hemisection, contusion, transection, and segment resection) were produced in male Sprague–Dawley rats to determine the most suitable animal model of SCI by analyzing the changes in diffusion tensor imaging (DTI) parameters both qualitatively and quantitatively in vivo. Radiological examinations were performed before surgery and weekly within 4 weeks after surgery to obtain DTI tractography, MRI routine images, and DTI data of fractional anisotropy (FA) and apparent diffusion coefficient (ADC). The Basso, Beattie, and Bresnahan scale was used to evaluate the locomotor outcomes. We found that DTI tractography tracked nerve fibers and showed conspicuous changes in the injured spinal cord in all the model groups, which confirmed that our modeling was successful. A decrease in FA values and an increase in ADC were observed in all the model groups after surgery. There were significant differences in FA and ADC between weeks 1 and 4 in both hemisection and contusion groups (P<0.05), whereas the differences in the transection and segment resection groups were not as remarkable (P>0.05). Basso, Beattie, and Bresnahan scores further proved the results because of a significant, positive correlation of the scores with FA (R=0.899, P<0.01) and a significant, negative correlation of the scores with ADC (R=−0.829, P<0.01). Therefore, the transection model, which is more quantified and stable within 4 weeks after injury according to the DTI and behavioral evaluation, should be used as the standard model for SCI animal testing.