A procedure for selecting reference areas in epidemiologic research employing census data and squared Euclidian distance is described. The procedure was adapted from cluster analysis, a multivariate statistical technique that has been applied in many disciplines. With the use of 12 census variables as the basis for evaluating sociodemographic differentiation, squared Euclidian distances were calculated between a geographically delineated index area in southwest Alberta, where residents had complained for several years about the effects of exposure to sour gas emissions, and 119 provincial census tracts in the rest of nonmetropolitan southern Alberta. The Euclidian distances can be interpreted as social distance scores with values close to zero representing a high level of sociodemographic similarity between the index area and potential reference areas. The social distance scores, in association with environmental data, suggested a clear choice for the most comparable unexposed reference area and illustrated the difficulty of finding a suitable most comparable exposed reference area. Results from the demographic component of the subsequent health survey indicated that the index area and reference area were similar in most respects. Furthermore, tests with and without statistical adjustment for confounding variables produced negligible differences on most of the important target outcome variables.