Convection-enhanced delivery (CED) is a novel drug delivery technique that uses positive infusion pressure to deliver therapeutic agents directly into the interstitial spaces of the brain. Despite the promise of CED, clinical trials have demonstrated that target-tissue anatomy and patient-specific physiology play a major role in drug distribution using this technique. In this study, we retrospectively tested the ability of a software algorithm using MR diffusion tensor imaging to predict patient-specific drug distributions by CED. A tumor-targeted cytotoxin, cintredekin besudotox (interleukin 13-PE38QQR), was coinfused with iodine 123-labeled human serum albumin (123I-HSA), in patients with recurrent malignant gliomas. The spatial distribution of 123I-HSA was then compared to a drug distribution simulation provided by the software algorithm. The algorithm had a high sensitivity (71.4%) and specificity (100%) for identifying the high proportion (7 of 14) of catheter trajectories that failed to deliver drug into the desired anatomical region (p = 0.021). This usually occurred when catheter trajectories crossed deep sulci, resulting in leak of the infusate into the subarachnoid cerebrospinal fluid space. The mean concordance of the volume of distribution at the 50% isodose level between the actual 123I-HSA distribution and simulation was 65.75% (95% confidence interval [CI], 52.0%-79.5%), and the mean maximal inplane deviation was less than 8.5 mm (95% CI, 4.0-13.0 mm). The use of this simulation algorithm was considered clinically useful in 84.6% of catheters. Routine use of this algorithm, and its further developments, should improve prospective selection of catheter trajectories, and thereby improve the efficacy of drugs delivered by this promising technique.