In this paper we propose a model for constructing neighbourhoods based on geo-referenced data and administrative data. The 431,233 inhabited hectare cells in Denmark are clustered into 9,404 small and 2,296 large neighbourhoods, inhabited on average in 2004 by 572 and 2,343 persons respectively. The priorities in the clustering process are to obtain neighbourhoods that are unaltered over time, delineated by physical barriers, compact, homogeneous in terms of type of housing and ownership, relatively small, homogeneous in terms of number of inhabitants, and comprised of a contiguous cluster of cells. To illustrate the importance of detailed neighbourhood information we compare social and ethnic segregation measured by Isolation and Dissimilation indices on the levels of municipalities and of small neighbourhoods. Our findings demonstrate substantial variation in the residential mix in neighbourhoods within a given municipality, and thus show the importance of having information on a more detailed geographical level than that of the municipality.