Our objective was to evaluate human immunodeficiency virus (HIV) risk-factor assessment and counseling behavior in 86 medical house staff and to ascertain the effect of question format (closed versus open) on reported physician behaviors. We designed a cross-sectional survey using a self-report questionnaire; we received 78 returns of 86 questionnaires (91% response) from one-year and three-year medical housestaff at two general medical clinics in a university-affiliated Bronx municipal hospital. House staff reported assessing sexual behavior (51%) and intravenous drug use (81%) in new patients significantly less often than they assessed smoking (95%) or alcohol use (94%). Counseling to reduce the risk of HIV transmission among all patients occurred even less frequently (41%, all P < .05). Question format (closed versus open) significantly affected response rate; respondents to closed-format questions reported far more behaviors performed. Intravenous drug users (IVDUs) received more frequent advice than general clinic patients about reducing HIV transmission (85% versus 41%), but house staff may ignore sexual risks in IVDUs. Resident confidence that IVDUs would change sexual or drug risk behavior was equally low, 1.5 on a 4-point scale. They reported greater discomfort discussing sexuality than drug use (2.4 versus 1.7; 4-point scale, P = .003). Discomfort discussing sexuality inversely associated with sexual history-taking in multi-variate analysis. Greater involvement with HIV positive patients and fewer perceived barriers to counseling were also associated with more assessment or counseling about risks for HIV transmission. These results suggest that enhancing house staff assessment and counseling efforts to reduce risks for HIV transmission requires interventions that foster involvement with HIV positive patients, emphasize the importance of sexual transmission in IVDUs, facilitate competence in discussing sexual behavior, and address provider belief that IVDUs will not change risk behaviors. Results also demonstrate a significant effect of question format on reported physician behavior.