Abstract: The false water level,which caused by the change of the number of steam bubbles which follows the change of local pressure due to the change of steam flow rate,make the nuclear steam generator water level difficult to identify.In order to improve the effect of identification,the identification method based on neural networks for nuclear steam generator water level is researched in this paper.The series-parallel model is applied to the identification and the back propagation algorithm of Levenberg— Marququardt type（LMBP） is employed to train the network.The proposed method with a good performance of identification is demonstrated by simulation results.