Dialect atlases comprise considerable numbers of linguistic feature maps, i.e. dialect maps representing one linguistic feature each. Large amounts of data like these are often difficult to handle. This article presents a new quantitative method for the automatic analysis of such large corpora of linguistic feature maps. It makes use of geographical similarities between single maps to establish a system of criteria for structural relatedness. Furthermore, it employs statistical techniques to test whether given linguistic relations between the maps coincide significantly with structural relations. To achieve this, each underlying point-symbol map is converted into an area-class map (with all the original information still available). These area-class maps yield additional information regarding their structural composition. Cluster analysis is then employed to obtain groupings of similar maps. Such groupings facilitate the search for language-internal factors that influence the geographical distribution of linguistic variants, as the relevance of any given linguistic parameter for spatial patterns can be tested using statistical methods. Moreover, language-external factors, such as topographical conditions, can be tested in the same way. Thus, this new method allows for a profound and substantiated investigation of the regularities that can be found in the geographical distributions of linguistic variants.