This paper proposes a RiskRank method to analyze the network risk propagation by studying the path and law of the network risk propagation. By computing the similarity and proximity between network nodes, a graph of network risk propagation is built, based on which a network risk propagation model is trained by iterations of random walk. The model is used to dynamically analyze the process of network risk propagation and quantitively evaluate the risk of network nodes. A high-risk clustering method is proposed based on the density clustering algorithm to isolate the high-risk area, thus controlling the security risk. The experimental results show that the accuracy, the precision and the recall of the RiskRank model is 97.4%, 98.1% and 86.4%, respectively.