Traumatic brain injury (TBI) commonly results in residual memory difficulties. Such deficits are amenable to cognitive rehabilitation, but optimal selection of rehabilitation interventions remains a challenge. We hypothesized that diffusion tensor imaging (DTI) could be used to predict which individuals were likely to benefit from a specific memory rehabilitation intervention. Thirty-seven individuals with TBI, of all severities, first underwent DTI scanning, along with 18 matched controls. Participants with TBI then attended a 12-session memory intervention emphasizing internal memory strategies (I-MEMS). Primary outcome measures (HVLT, RBMT) were collected at the time of DTI scanning, and both immediately and one month post-therapy. In contrast to typical neuroimaging analysis, fractional anisotropy (FA) was used to predict long-term outcome scores, adjusting for typical predictors (injury severity, age, education, time since injury, pretest score). FA of the parahippocampal white matter was a significant negative predictor of HVLT, while the anterior corpus callosum, left anterior internal capsule, and right anterior corona radiata were negative predictors of RBMT outcome. The importance of these predictors rivaled those of pretest scores. Thus, FA measures may provide substantial predictive value for other cognitive interventions as well. The reason why higher FA was associated with less successful response to cognitive intervention remains unclear and will require further study.