The social presence of a human interviewer can affect respondents’ answers to sensitive questions, causing them to overreport socially desirable behaviors and underreport socially undesirable ones. Self-administration methods such as ACASI can mitigate some problems associated with interviewer-administration, but human interviewers have been shown to be better at eliciting certain kinds of information from respondents, as well as increasing respondents’ willingness to participate in and complete a survey. The study reported here explores whether new technologies for human-looking virtual interviewers could potentially combine the benefits of both self-administration and interviewer-administration. In a laboratory experiment, 240 respondents answered threatening questions either with a human interviewer face to face (FTF), or by clicking in a web browser with one of three self-administered modes: a virtual interviewer designed to include a high number of social cues (i.e. facial movements and expressions), a virtual interviewer with minimal social cues, or via ACASI. The results indicate that the number of social cues present in the virtual interviewers interact with the type and degree of question sensitivity. For example, when asking respondents to report their number of lifetime sex partners, the highly expressive virtual interviewer elicited answers similar to what was seen in the FTF interview (fewer partners) while the minimally expressive virtual interviewer elicited answers similar to what was found with ACASI (more partners). However, when respondents were asked to describe their weight, which could be visually confirmed by a human interviewer, respondents interacting with the highly expressive virtual interviewer described their weight as lower than those who participated in the FTF interview, presumably understanding that the virtual interviewer couldn’t see them. Overall, the results indicate that the virtual interviewers sometimes produce social presence effects and sometimes don’t, depending on the way in which, and the degree to which, a given question is sensitive.