An on-line identification method for switching nonlinear systems with multiple Hammerstein-Wiener models based on neural networks is proposed. First of all, a switching rule prediction model for switching non-linear systems is established through a BP neural network; Secondly, the recursive method with discounted measurements is proposed to identify the parameters of each nonlinear subsystem. The key term separation method is used to separate the product terms, and the parameter estimates of each subsystem are obtained. Finally, an example of nonlinear system identification is simulated and compared with other methods to verify the effectiveness of the proposed method.