Abstract: Kernel k-means is an extension of the standard k-means that improves the quality of nonlinearly separable clusters.In order to overcome the cluster initialization problem such as long computation time and low efficiency with this method, In this paper we propose theD²weight kernel k-means algorithm.Our method adds one cluster depending on the contribution to the sun of the distance that any point to the closest center that we have already chosen, and we choose the point as the cluster that has the most contribution.In order to verify the effectiveness of the proposed algorithm, we have done a large number of experiments on the UCI dataset.The experiment result shows that the proposed algorithm improves both the accuracy and the speed of kernel k-means.The new algorithm does better than the traditional k-means.