The aim is to maximize between-group variance and to minimize within-group variance. It starts with each case in a separate cluster and that combines the clusters sequentially, reducing the number of clusters at each step until only one cluster is left.