Proper and rapid identification of malfunctions is of premier importance for the safe operation of Nuclear Power Plants (NPP). Many monitoring or/and diagnosis methodologies based on artificial and computational intelligence have been proposed to aid operator to understand system problems, perform trouble-shooting action and reduce human error under serious pressure. However, because no single method is adequate to handle all requirements for diagnosis system, hybrid approaches where different methods work in conjunction to solve parts of the problem interest researchers greatly. In this study, Multilevel Flow Models (MFM) and Artificial Neural Network (ANN) are proposed and employed to develop a fault diagnosis system, with the intention of improving the successful rate of identification on the one hand, and improving the understandability of diagnostic process and results on the other hand. Several simulation cases were conducted from a newly developed full-scale NPP simulator for evaluating the performance of the proposed hybrid approach. This paper will introduce the proposed hybrid approach, simulation experiment and its results.