OBJECTIVE:To assess the association between protease inhibitor (PI) use and the incidence of diabetes mellitus (DM) among participants in the Women's Interagency HIV Study. DESIGN:Prospective multicenter cohort study. The diagnosis of DM was based on self-report at semiannual interviews conducted from 1994 to 1998. SETTING:Six inner-city clinical sites in the United States (Brooklyn, NY; Bronx, NY; Washington, DC; Chicago, IL; San Francisco, CA; and Los Angeles, CA). PARTICIPANTS:A total of 1785 nonpregnant women who had no history of prior DM. The women made up four groups1) PI users (n = 609, person-years PY at risk = 707); 2) reverse transcriptase inhibitor (RTI)-only users (n = 932, PY = 1486); 3) HIV-infected women reporting no antiretroviral therapy (ART) ever (n = 816, PY = 1480)and 4) HIV-uninfected women (n = 350, PY = 905). MAIN OUTCOMES:Incidence of DM and median body mass index (BMI) from 1995 to 1998 were compared among the four groups. RESULTS:Sixty-nine incident cases of DM occurred among 1785 women (1.5 cases per 100 PY; 95% CI1.2-1.9). The incidence of DM among PI users was 2.8 cases per 100 PY (2.8%) versus 1.2% among both RTI users and women on no ART (95% CI1.6-4.1 PI; 0.7-1.8 RTI and no ART; P = 0.01 for comparison of the PI group with the RTI group) and 1.4% among HIV-uninfected women (95% CI0.7-2.2, P = 0.06 for comparison with PI group). Weight gain was not associated with either PI or RTI use. Multivariate models identified PI use (hazard ratio HR = 2.90 95% CI1.50-5.60; P = 0.002), age (HR = 1.75 per 10 years 95% CI1.31-2.34; P = 0.0002) and BMI as independent risk factors for DM. CONCLUSIONS:PI use was associated with a threefold increase in the risk of reporting incident DM. Routine screening for diabetes, particularly among older and heavier patients using PI therapy, is advisable.