Abstract: By on-line monitoring the equipments in nuclear power plant (NPP), their working conditions could be known timely and the faults could be predicted, so that the safety and economy of NPP can be improved. The monitoring and diagnosis of bearing faults in rotatory machines in NPP were studied. The energy operator method was adopted for the envelope demodulation of the vibration signals, which could find the fault characteristic frequencies from the noises and modulating signals. To automatically determine the center frequency of the band-pass filter, the frequency band variance was proposed as an index, which achieved the maximum around the resonant frequency. The feasibility and validity of the proposed method were verified by the experimental data. The results show that the conditions of the bearings are effectively monitored, and fault types are correctly diagnosed.