Using a survey of New York State Residence-Trained Family Physicians and the 1990 census data, this paper assesses the relative importance and consistency of factors associated with physician practice locations when different definitions of community size are used. By matching the zip code information with 434 physicians' practice locations, physician respondents' self-reported communities are linked to census-defined communities. It was found that the significant level of some variables could be affected when community classifications were based on survey responses rather than census data. It concludes that caution should be taken for interpreting rural-urban differences when data are solely based on self-reported practice locations.