Survey interviewers inevitably display non-verbal behavior (e.g. nods, smiles, eye gaze), which can be observed by respondents and may affect their answers. While many technological innovations have reduced the interviewer's presence in data collection (telephone eliminates visual information, ACA5I eliminates the interviewer for sensitive questions, web questionnaires eliminate the interviewer altogether), one new technology - embodied conversational agents (ECAs) - promises to bring many features of human interviewers to automated data collection. ECAs are graphical projections of full-bodied humans in the computer interface and can fall anywhere from between absolute realism and to cartoon depiction. We have explored ECAs In a research program that is now over a decade old. Because ECAs are conversational, i.e. speak, understand speech, and acknowledge their understanding, they are plausible interfaces for automated survey interviews. In this paper we consider ECAs as potential interviewers but first look lo the survey literature on observable attributes of interviewers and interviewer rapport to guide our discussion of ECAs. While there are well-documented effects of attributes such as race, gender and age of interviewer, the results are more complex than often assumed and the exact mechanism behind such effects is never articulated. There is ambiguity about the value of rapport between interviewers and respondents, but for better or worse rapport derives from interviewers' non-verbal behavior. We (Cassell, Bickmore and colleagues) have examined rapport and small talk in ECAs and have found that this fosters a sense of trust in some users (extroverts) relative to ECAs that are more businesslike. In fact the general pattern that has emerged is that people treat ECAs much as they treat humans. Thus while ECAs ore promising for training interviewers and experimentally exploring the mechanisms behind rapport, they may be too human-like to create the professional distance that survey data collection requires, especially for sensitive questions.