In recent years, researchers and policymakers have used data from large-scale surveys of physicians to address important issues. A review of several of these surveys explores potential problems in this method of gathering data on physicians' services. To obtain a better grasp of the limitations such problems may pose, we examine several recent surveys, comparing response rates and survey findings, and in one survey the reliability of individual items. Response rates appear highly sensitive to differences in the approaches made to respondents and their perceptions of the goals of individual investigations. Reliability of survey items seems to depend on the specificity of information requested. Variation in the findings from different surveys may occur for many reasons, but is most likely to be found in response to items whose presentation differs in each survey's research instrument. Data from these surveys appear clearly useful for some important purposes. The large-scale medical practice survey seems particularly valuable in generating an understanding of differences among specialties in resources used in the delivery of care. Nevertheless, researchers and policymakers must understand the steps necessary to obtain reliable results and possible limitations in the accuracy of findings to make the best use of survey methodology as applied to medical practice.