The purpose of this study was to assess the reliability of responses to HIV risk behavior questions obtained using a voice-enhanced, computer-administered self-interview (audio-CASI) system with touch-screen response compared with those obtained via face-to-face interviews administered by trained and experienced interviewers. Additionally, the study assessed bias that may be attributable to an audio-CASI data collection format. The design of the study was a four-group crossover design with random assignment to one of four study conditions: (a) audio-CASI interview at both intake and retest, (b) face-to-face interview at both intake and retest, (c) audio-CASI interview at intake and face-to-face interview at retest, and (d) face-to-face interview at intake and audio-CASI interview at retest. The study was conducted with a sample of drug users at risk for HIV infection interviewed in nonclinical settings. Data were collected at intake and 48 hours after intake. Analyses show that data obtained using voice-enhanced computer interviewing with touch-screen response are reliable and are comparable to data obtained using interviewer administered face-to-face interviews. However, bias was found to be associated with data collection format and may be partially attributable to the complexity of the questionnaire.