In this paper, we aim to solve the problem of neural network in computer network security assessment, which is very important for computer network's popularization. Our proposed computer network security assessment system contains client and server. The client module includes: 1) scanning configuration model, 2) assessment model, 3) scanning result database model and 4) output model. Furthermore, server is made up of 1) scanning engine, 2) vulnerability database, 3) rules database. To promote the performance of artificial neural network, we choose the back propagation neural network, and particle swarm optimization is utilized to optimize parameters. Finally, experimental results demonstrate the effectiveness of our proposed approach.