Due to rapidly increasing complex attacks, networks become more and more insecure. How to accurately predict the future security situation of networks is thus an important research issue. Forecasting security situation can improve the awareness of network states and provide decision support to threat analysis and network planning. This paper provides a combination model of neural networks to predict the security situation of computer networks. Our contribution is in two aspects. On the one hand, we select several single neural network models including Backward Propagation (BP) network, Elman network, and Radial Basis Function (RBF) network to construct the combination model. On the other hand, we use the entropy method to determine the weights of each single model in the combination model. Experimental results show that the proposed combination model can predict the security situation of networks more e?ectively than any single neural network.