Abstract: Direct Generalized Predictive Control（GPC） based on Elman neural network for a class of nonlinear system with unknown parameters is presented to overcome the high load of computing of traditional GPC as online recursion of Diophantine equations and matrix inversion.In this method，the nonlinear system is substituted with a time varying linear system，then the dynamic modeling error is estimated，and the controller parameter vector θ is adjusted adaptively，finally an Elman neural network is used to approximate the function of control increment.It is proved that the proposed method can make the tracking error converge to a little neighborhood of the origin.Simulation results demonstrate the effectiveness of this proposed method.