Background.— To determine the incidence of agranulocytosis, a descriptive epidemiologic study was performed. Methods.— With the use of computerized Medicaid billing data from 1980 through 1985 from Minnesota, Michigan, and Florida, the ratio of persons hospitalized with a discharge diagnosis of neutropenia to persons with any claim for medical service was first used as an estimate of the incidence rate of the condition. Patients with cancer and patients receiving cytotoxic and immunosuppressive drugs were excluded. The information provided by a review of medical records for a subset of neutropenia cases was used to determine the proportion with neutropenia after excluding cases with recurrent or chronic neutropenia, and to determine the proportion with agranulocytosis. Results.— The incidence rates (95% confidence intervals) of agranulocytosis, excluding recurrent or chronic disease, were 2.3 (1.4 to 3.7), 7.7(6.6 to 8.9), and 15.4 (11.3 to 20.4) per million per year in each state, respectively. The overall incidence was 7.2 (6.3 to 8.1) per million per year. Conclusions.— Agranulocytosis is an extremely uncommon condition. The excess risk of agranulocytosis due to any drug other than cytotoxic drugs must, therefore, be very low. (Arch Intern Med. 1992;152:1475-1480)