The paper puts forward a novel realtime monitoring and fault diagnosis system for a printing machine’svibration based on discrete wavelet transform (DWT) and probabilistic neural networks(PNN). With DWT, the printing process sig nal can be diagnosed, and some component in special frequency band is selected to reconstruct. With the decomposition and reconstruction of wavelet packet, the monocomponents with fault feature in different frequency bands are captured and separated out. For these symptom parameters, the neural network mapping is empl oyed to diagnose the printing machine faults further. Simulation results show th at the proposed method is highly efficient, accurate and easy in practical appli cation.