Abstract: According to defects of traditional performance evaluation models, BP neural net work is used to evaluate performance of capital operation. Some optimizations to BP neural networks are done. The learning mechanism of self adaptation optimize s the learning ratio of BP neural networks. The accelerating mechanism of moment um item accelerates the constringent velocities of BP neural networks. The genet ic algorithm is used to select initial parameters of BP neural networks. Lastly, performance data of capital operation in 500 industrial enterprises is used to prove validity of the model.