Abstract: The initial clustering centers of traditional K-means clustering algorithm are generated randomly.Different initial clustering centers will lead to different clustering results.It means that unstable results were often obtained owing to random selection of initial centers while using traditional K-means.A new method about optimization of initial centers is brought forward based on K-means clustering algorithm which can gain high accurate results.The main idea of the improved algorithm is to choose the K-means initial clustering centers through calculating the local outlier factor of all sample data.The greatest distance dense points are chosen to eliminate the influence of local outliers.The experimental results show that the improved algorithm can reduce the selection sensitivity of the K-means initial clustering centers,and also get more accurate clustering results with less iterations.