Aiming to the shortage of BP neural network in training algorithm, the problem of neural network learning can be seen as a function optimization problem and the neural network model based on two order particle swarm optimization is proposed. Then, chaotic system is identified by BP trained with two-order PSO and the efficiency of BP trained with two-order PSO is compared with those of BP and RBF based on the identification of chaotic system. The experimental results show that BP trained with two-order PSO is better than BP and RBF used in chaotic system identification.