To solve the problem of vibration signal denoising， a stacked denoising auto-encoder method is proposed. Combined with PReLU activation function and Batchnormalization ， the traditional stacked denoising auto-encoder is improved to enhance the capability of feature extraction and signal reconstruction. The stacked denoising auto-encoder method uses the encoder to extract the features of the vibration signal with noise and the decoder to reconstruct the signal， so as to realize the denoising of the vibration signal. Denoising experiments are carried out under the conditions of sinusoidal signal， amplitude modulation signal and simulated bearing fault signal. Tested results are better than those of traditional noise reduction methods. Finally， measured bearing vibration signals are used in the experiment. Under the condition of strong noise， bearing vibration signals added with noise can be effectively reduced. Experimental results show that the proposed stacked denoising auto-encoder method can be applied to accomplish vibration signal denoising.