Momentum wheel is a key actuator of the attitude controlling system of the satellite. Thus, it is very meaningful to accomplish its fault diagnosability to maintain the working order of the satellite. Therefore, the operator must be timely to judge change-trend in momentum wheel, figure out the reasons for changing conditions and adjust it promptly. However, the working environment of the momentum wheel is very complicated, targeting on the problems of complex fault mechanisms and uncertainty between fault type and fault symptoms, this paper puts forward a Bayesian network structure based on the FMEA (Failure Modes and Effects Analysis) Bayesian diagnostic network. This method will provide theoretical basis for momentum wheel fault diagnosis and the simulation shows it can detect the fault by detecting multi-fault variables effectively.