A mathematical method is proposed for clustering a set of elements with respect to a set of criteria. Input information is described in terms of a graph representing a distance function between the elements. Definitions are given for the concepts of a ‘tree’, and a ‘tree‐representation’. It is shown that clusters are particular kinds of subsets (level‐subsets) in a tree‐representation; and that a final pattern contains at least one non‐trivial tree, under specified conditions. An example of clustering is set out, and the computation of level‐subsets is described. The selection of clusters from a final pattern is