It is difficult to describe hydraulic turbine generating unit system via accurate mathematics model because it is a complicated non-liner system. In the paper, RBF neural networks models were established to identify hydraulic turbine generating unit. In RBF networks training, a practical learning algorithm was proposed for adjusting effectively the node number, centers and width of Gaussian function of hidden layer nodes. Off-line training and on-line identifying were combined together to train networks and identify hydraulic turbine generating unit system. Simulation results show that the designed model can well identify the characteristic of hydraulic turbine generating unit. Thus, the identifying model can lay the good foundation for study on intelligent control strategies of hydraulic turbine generating system.