Abstract: Size constraints can improve the clustering performance of clustering methods.However the differences in the size of clusters,i.e.the number of instances contained in each cluster will decrease the clustering performance.This paper introduces a new scheme of size constraints on size of each cluster and transforms them into linear programming optimization.Experiments results on UCI benchmark datasets show that the new method outperforms the random scheme.The clustering performance can be increased even when the size constraints are relaxed to some extent.The new algorithm can increase the clustering accuracy efficiently.