The likelihood that subjects in human immunodeficiency virus (HIV) vaccine efficacy trials will alter their behavioral risks for HIV infection over time must be considered in evaluating the feasibility of such trials and in estimating the necessary sample sizes to be enrolled. Potential subjects for future vaccine efficacy trials include injecting drug users (IDUs) and others who may be difficult to retain in studies and who may alter HIV-risk-related behaviors substantially over time. We have investigated behavior change, retention, and HIV seroconversion among 577 New York City resident IDUs and sexual partners of IDUs enlisted between July 1 and December 31, 1992. We attempted to see all subjects every 3 months for interviews, blood donation and HIV testing. We were able to retain 68% of subjects in the study through the third scheduled recall at 7.5-10.5 months after enlistment. HIV seroconversion through March 1, 1994, was 1.33/100 person-years at risk. There was a significant inverse relationship between HIV seroconversion and retention at the 9-month recall after adjusting for age, gender, and the amount of locator information provided by subjects at enlistment. Among subjects seen at each of the scheduled visits at 3, 6, and 9 months after enrollment, modest but statistically significant behavior changes that reduced risk were observed in self-reported drug injection frequency, heroin injection frequency, sexual contact with IDUs, and sharing of needles/syringes. The magnitude of these changes in risk, however, was small and may be transient. The behavior changes observed to date do not appear to be large enough to substantially alter calculations of sample sizes needed in future HIV vaccine efficacy trials.