On-line panels of volunteer respondents have emerged as a new method of conducting surveys for market and public opinion research with substantial cost and logistical advantages over traditional mail or telephone surveys. However, because they are not based on probability sampling, the results from on-line panels raise serious concerns about their validity in terms of representing the characteristics or views of the population. With support from the Alfred P. Sloan Foundation, the CivilPanel (formerly eTownPanel) project was created to test the validity of an on-line panel approach to citizen surveys about government performance. This article reports on the development of the project, including the growth and composition of the panel, and on the validity of its on-line survey results as defined by comparisons with established, randomsample surveys of public opinion. Implications of the findings for the practice of citizen surveys as well as for the study of public opinion about local government performance are discussed.