Objective: To examine performance of models in predicting health care utilization for individuals with spinal cord dysfunction. Design: Regression models compared 2 diagnosis-based risk-adjustment methods, the adjusted clinical groups (ACGs) and diagnostic cost groups (DCGs). To improve prediction, we added to our model: (1) spinal cord dysfunction-specific diagnostic information, (2) limitations in self-care function, and (3) both 1 and 2. Setting: Models were replicated in 3 populations. Participants: Samples from 3 populations: (1) 40% of veterans using Veterans Health Administration services in fiscal year 1997 (FY97) (N = 1,046,803), (2) veteran sample with spinal cord dysfunction identified by codes from the International Statistical Classification of Diseases, 9th Revision, Clinical Modifications (N = 7666), and (3) veteran sample identified in Veterans Affairs Spinal Cord Dysfunction Registry (N=5888). Interventions: Not applicable. Main Outcome Measures: Inpatient, outpatient, and total days of care in FY97.2 Results: The DCG models (R-2 range, .22-.38) performed better than ACG models (R-2 range, .04-.34) for all outcomes. Spinal cord dysfunction-specific diagnostic information improved prediction more in the ACG model than in the DCG model (R-2 range for ACG, .14-.34; R-2 range for DCG, .24.38). Information on self-care function slightly improved performance (R-2 range increased from 0 to .04). Conclusions: The DCG risk-adjustment models predicted health care utilization better than ACG models. ACG model prediction was improved by adding information.