Clustering of distributions

Does anybody know of any methods or modifications of methods which could define clusters of multinomial distributions? For example, imagine developing a group of N race/ethnicity distributions based on a dataset of all census tracts.