The computerized outpatient records of the Harvard Community Health Plan, a 230,000-member health maintenance organization, were used to determine the frequency with which serum sickness is recognized in the practice setting after exposure to antibiotics. The medical records of 3,487 children who had been prescribed cefaclor or amoxicillin were searched in December 1986 for coded diagnoses of serum sickness and related conditions. Diagnoses were validated by blinded review of dictated and written office notes. There were 12 cases of serum sickness in 11,523 child-years. During this time, these children were prescribed 13,487 courses of amoxicillin, 5,597 courses of trimethoprim-sulfamethoxazole (TMP-SMZ), 3,553 courses of cefaclor, and 2,325 courses of penicillin V. Serum sickness was considered to be antibiotic-related if it occurred within 20 days of initiation of antibiotic therapy. Five cases were temporally associated with cefaclor, one with both amoxicillin and TMP-SMZ, four with TMP-SMZ alone, and one with penicillin V alone. One case was not associated with any antibiotic exposure. All antibiotic-related cases occurred in children under age 6 years who were treated for otitis media or streptococcal pharyngitis, and most cases began 7-11 days after initiation of antibiotic. All but one of the antibiotic-related cases occurred in children who had relatively heavy lifetime antibiotic exposure. The risk of serum sickness was significantly elevated after cefaclor compared with amoxicillin, even among the most heavily exposed children (relative risk = 14.8, p = 0.01, 95% confidence interval 2.0-352.0). Most cases prompted several physician visits, but none required hospitalization.