OBJECTIVE Reported estimates of the risk of agranulocytosis associated with metamizol have varied by several orders of magnitude. We assessed this association in a large database for the surveillance of blood dyscrasias. METHODS Since 1980, all laboratory units of haematology in a defined area (3.3-4.1x10(6) inhabitants) contribute to the ascertainment of all cases of agranulocytosis meeting strict diagnostic criteria. These cases of patients with agranulocytosis and sex-, age-, hospital- and date-matched controls were interviewed using a structured questionnaire about previous drug exposures, and relative risks were calculated for several categories of exposure to metamizol. RESULTS After a total follow-up of 78.73x10(6) person-years, 273 community cases of agranulocytosis had been found--of which 96 were excluded for various reasons and 177 were included in the case-control analysis--and were compared with 586 matched controls. Thirty cases of agranulocytosis (16.9%) and nine controls (1.5%) had been exposed to metamizol during the week before the index day. The adjusted relative risk was 25.8 [95% confidence interval (CI), 8.4-79.1], and the attributable incidence was 0.56 (0.4-0.8) cases per million inhabitants and per year. The risk disappeared after more than 10 days since the last dose of metamizol, and it increased with duration of use. Those with agranulocytosis exposed to metamizol had taken the drug for longer periods than the exposed controls. Compared with the cases recently reported from Sweden, the duration of use of metamizol by our exposed cases was substantially shorter, and the use of concomitant medications potentially causing agranulocytosis was lower. CONCLUSIONS In our milieu, agranulocytosis attributable to metamizol is rare. Geographical disparities in its risk estimate can be partly explained by differences in its patterns of use, in terms of dose, duration and concomitant medications.