This article presents a methodology for the clustering of a large number of tables of similar form. Data of this kind is often available from National Statistical Offices as tabulations of a set of variables for each of a large number of geographic units. Model-based clustering is employed, the underlying model being a finite mixture of multinomial distributions. We discuss the choice of the number of clusters and the interpretation of the clusters found. As a case study we consider a set of age-group by sex tables for small areas obtained in the New Zealand Census of Population and Dwellings for the year 2006.