Beginning in Europe in the 1980s (Mantel in France and Willem Saris in the Netherlands}, and introduced by lnterSurvey from Knowledge Networks) in the USA in 1999, there have been a small number of attempts to recruit survey panels based on 'representative' samples of households, provide the households with computer hardware and a communication system, and then collect survey data from the households using remote data collection. While the strength of the concept is clear, the challenge has been to achieve a response rate consistent with the demands of 'influential' social and policy research. These challenges arise at the initial recruitment stage and at the subsequent web survey response stages. In 2006, the Science Resources Statistics division of the National Science Foundation (NSF) awarded a grant for a demonstration project to investigate whether households can be recruited to an internet panel combining area probability sampling, in-person recruitment for panel households, and web-based data collection. The ultimate aim was to establish whether there is a realistic prospect that this could be a cost-effective way of collecting data from high quality samples. The paper presents the results of the project. Response rates varied between 59% and 69% for three different approaches to recruitment. The presentation will investigate the impact of different aspects of the field approach and draw tentative conclusions about prospects for scaling the operation up to a more relevant size. In addition, using characteristics of the recruited households, predictors of panel participation will be examined. NORC carried out the sampling and recruitment, and Knowledge Networks was responsible for the web-related aspects of the data collection, including panel maintenance. Recruitment fieldwork was carried out in the summer of 2006.