Abstract: The problem of how to identify the nonlinear dynamic process using multilayer feedforward NN, trained by nonlinear BP algorithm, is investigated. The procedure of the identification method based on NBP networks is also given. Simulation results show that the performance of the model obtained with NBP network is at least as good as the normal BP network. Since NBP algorithm avoids computations of the derivative of neurons' activations, the results can be utilized to design NN based controllers at less cost than those based on normal BP.