This paper presents the neural network model prediction control (NNMPC) for the Francis hydroturbine generator set （FTGS） possessing nonlinear characteristics. The neural network identification model (NNIM) is used to predict future response to potential control signals of the FTGS. An optimization algorithm can compute the control signals that optimize future FTGS performance. Simulated results show that NNMPC is an effective tool for the nonlinear FTGS.