This study examines the ability of MedisGroups, a severity measure based on clinical data abstracted from the medical record, to predict hospitalization charges. MedisGroups measures severity both on admission and approximately 1 week into the hospital stay. The data base contained 23,361 admissions of Medicare beneficiaries in six conditions from 836 hospitals in seven states between January 1985, and May 1986. In all six conditions, higher admission and mid-stay severity scores were generally associated with higher charges. Across the six conditions, the R2 values for predicting charges using diagnosis-related group (DRG) class ranged from 0.06 to 0.32 using trimmed data. Adding admission MedisGroups scores to DRG class produced R2 values ranging from 0.09 to 0.33, while adding mid-stay scores yielded R2 values from 0.15 to 0.41, and adding both admission and mid-stay scores produced R2 levels ranging from 0.17 to 0.42. Very little of the superior predictive power of the mid-stay score could be attributed to its serving as a proxy for length of stay.