OBJECTIVES: To evaluate the accuracy of claims versus original prescriptions by calculating the match rate for 7 fields: drug, strength, dosage form, quantity, days supply, daily dose, and date dispensed. Claims accuracy was also analyzed by therapeutic class for antibiotics, oral hypoglycemics, H2 antagonists, antipsychotics, and antidepressants. METHODS: Purdue s Pharmaceutical Economics Research Center (PERC) systematically recruited 10 pharmacies from a random list generated by pharmacy benefits manager PCS Health Systems, Inc. (PCS). PCS randomly extracted 500 claims submitted by each pharmacy between 07/01/96 and 06/30/97. Pharmacies obscured patient information and photocopied the original prescriptions. Trained assistants abstracted 5 fields and calculated days supply and daily dose from the directions for use. RESULTS: Six of 10 pharmacies provided 2,938 photocopies, split approximately evenly among therapeutic classes. Overall match rates were: drug (99.9%), strength (97.0%), dosage form (95.5%), quantity (94.1%), days supply (85.2%), daily dose (83.2%), and date dispensed (65.1%). Applicable mean percentage differences were: strength (1.32%), daily dose (2.53%), days supply (2.73%), and quantity (6.03%). The mean difference for date dispensed was 19.3 days. Match rates were comparable for antidepressants, antipsychotics, and H2 antagonists. Antibiotic and oral hypoglycemic rates were more variable. CONCLUSIONS: Claims reliability was excellent for drug, strength, dosage form, and quantity and good for days supply and daily dose. Antidepressant, antipsychotic, and H2 antagonist classes had comparable reliability. The reliability of date dispensed appeared questionable, but the dates recorded by hand on prescriptions were later than those automatically submitted for claims, suggesting that claims data may be more accurate than the original prescriptions. This study found prescription claims to be a reliable data source for health care research.