OBJECTIVE: The objective of this study was to develop and validate a series of schizophrenia specific risk adjustment cost models. METHODS: Georgia Medicaid claims data linked with institutional inpatient data for 21,602 continuous eligible persons suffering from schizophrenia was used to build a prospective diagnosis-based, a demographic-based, a drug-based, and a combined risk adjustment cost model. ICD-9-CM and drug category classifications were derived from the literature and supplemented by an expert panel. Variables were screened and cost weights were derived empirically in a random 50% training sample using a robust a weighted Heuber-White regression model and validated by expert panel review, bootstrapping methods, and assessing indices of discrimination in a 50% validation sample. Model calibration and correlations of errors with policy relevant groups were also estimated. RESULTS: Measures of discrimination (R2) varied between 16.4% for the ICD-9-CM based model to 21.8% for the combined model for trimmed total cost and varied between 4.9% to 11.3% for mental health costs in the validation sample. Risk adjustment models based on drug or ICD-9-CM information discriminated costs equally well and the combined models outperformed both drug and ICD-9-CM based models. A simple model using prior year costs combined with demographic covariates had R2s > 40% for both mental health and total costs. CONCLUSIONS: The drug and ICD-9-CM based models performed equally well and either can be used with equal confidence depending on data availability. The combined models performed better than either the ICD-9-CM or drug based models indicating that drug exposure information can compliment more traditional approaches. Health services researchers wishing to control for differences in comorbidity and severity that influence cost should always consider including prior utilization (costs) since prior year costs were vastly superior predictors of costs.