The reliability of acquisition terminal software is an essential criterion for the life span of software system. In view of the difficulties of parameter optimization and low accuracy of software system predictive model existing in multi-neural network (NN) method and SVP method in reliability evaluation of software system, method of building model based on simulated annealing genetic fusion algorithm (SAGFA)-BPNN is proposed in this paper. Firstly, PCA is used to reduce the dimension of the experimental data, this is to eliminate the redundancy and interference sample that will affect the accuracy of model. Then, SAGFA is developed based on the optimization of SA and GA method. Furthermore, by utilizing the global optimization property of SAGFA method and the non-linear mapping property of BPNN, the SAGFA-BPNN and the method of its model building are given. This will raise the training speed, improve the ability of global optimization and increase the accuracy of prediction. This method is then applied to predict the reliability of acquisition terminal software and the prediction result shows that the proposed method can effectively raise the accuracy of model..