A novel global clustering method called the greedy elimination method is presented. Experiments show that the proposed method scores significantly lower clustering errors than the standard K-means over two benchmark and two application datasets, and it is efficient for handling large datasets.