We estimated HIV incidence among injection drug users attending a drug treatment clinic in San Francisco from 1995 to 1998 using two methods. An anonymous sequential testing method identified no seroconversions among clients seen more than once during the period (one-sided upper 95% confidence limit 1.02 per 100 person-years). A sensitive/less sensitive immunoassay testing strategy detected no early infections (one-sided upper 95% confidence limit 1.90% per year). Methods were concordant and feasible in the setting. Although detection of no new HIV infections in this population of injection drug users (IDUs) is encouraging, epidemiologic studies among IDUs not in treatment are needed to monitor the HIV epidemic effectively.