We report on difficulties in applying traditional clustering procedures to discrete data. We describe a graph theoretical approach in clustering binary vectors where the number of clusters is not given in advance. New clustering procedures are combined from several algorithms and heuristics from graph theory.