In order to explore a priori hypotheses about drug-induced Stevens-Johnson Syndrome (SJS), a case-ontrol study was initiated using data from COMPASSR, a computerized data base consisting of Medicaid claims data. The records of 3.8 million patients in five U.S. states were searched to identify patients with an inpatient diagnosis of ICD-9-M code 695.1 (erythema multiforme-EM). Out of the total of 367 cases that were identified, primary medical records for 249 were sought and 128 (51.4 per cent) of these were obtained. The remainder could not be obtained because: in 36 (29.8 per cent) the hospital refused to provide medical records; in 33 (27.3 per cent) there were transcription errors; in 20 (16.5 per cent) the state could not translate the identification number, primarily because the patients lost Medicaid eligibility too long before our request; in 27 (22.3 per cent) the hospital could not locate the patient's record; and in 5 (4.1 per cent) there were other reasons. Of those with a medical record, 121 (94.5 per cent) had a skin diagnosis and 109 (85.2 per cent) had a diagnosis compatible with ICD-9-CM code 695.1 specified on their discharge summary. However, in 35 (27.3 per cent) an expert reviewer felt that the discharge diagnosis was incorrect. In 50 (39 per cent) the computer diagnosis was incorrect. Only 19 (14.8 per cent) were judged by the expert reviewer to truly have Stevens-Johnson Syndrome, and an additional 37 (28.9 per cent) were judged to have erythema multiforme minor. Thus, the computerized diagnosis agreed very well with the diagnoses specified on the discharge summary. However, EM is frequently misdiagnosed, ICD-9-CM code 695.1 contains multiple other diagnoses which are not EM, and much of hospitalized EM is EM minor. Thus, studies of SJS cannot be performed except in patients whose medical records are available.