Considering the fact that conventional K-means algorithm is susceptible to the outliers and noise points,and lacking in robustness,a new K-means algorithm based on co-entropy is proposed. The proposed algorithm employs co-entropy as a means of local similarity measurement,and follows the co-entropy maximization principle to solve the optimal cluster centers. An iteratively reweighted optimization technique is employed to quickly find the optimal cluster centers. For outliers and noisy data points with larger residuals,they will be assigned smaller weights in updating the cluster centers. Experimental results demonstrate that the proposed co-entropy based K-means algorithm is robust,winning a better clustering effect.