The many modes of collecting survey data all have strengths and weaknesses. For example interviews are less expensive and lead to higher response rates when conducted on the phone thon face-to-face. However, telephone conversations involve only speech, which leaves no reviewable trace ('What were those response options, again?') and is highly sequential (I speaks, then R speaks, then I speaks, etc.). There is no way to display visual materials like images or lists of response options, and deaf respondents cannot be interviewed over the telephone. Web-based questionnaires, in contrast, often display text and graphics, can be completed at the respondents convenience and are not prone to interviewer influences. However, web respondents typically communicate by typing and clicking, an impoverished form of input relative to speech. The blind and those with motor disabilities are not able to easily respond to questions on the web. Multimodal interfaces, which I have been developing for nearly a decade, are a promising addition to the set of technologies currently for automating or supporting survey data collection. These recognize users' speech, eye gaze, gesture, writing and touch; they produce speech, graphical displays, and gestures (on screen or embodied in a conversational agent). They allow the user to choose an input modality, enabling users with various disabilities to use the same underlying system, and some systems also allow 'composite' input, e.g. when asked "How many bedrooms are there in your house?" o respondent might say "these rooms ore bedrooms" and circle the relevant rooms on a floor plan, allowing the system to determine the answer. Multimodal interfaces can improve automatic speech recognition by using gesture and other Input to clarify what was said so could help increase the practicality of speech interfaces for survey tasks. We discuss the applicability of multimodal interfaces to survey interviews and self-administration.