Abstract: Non-local mean filtering denoises the image based on the similarity between pixels in the image. However, this method cannot effectively suppress the abnormal amplitude noise in the image. The seismic signals simultaneously in common offset gathers (equal offset gathers sorted by shot coordinates) have the characteristics of high similarity and concentrated distribution. So, the abnormal amplitude noise in the signal can be judged according to the weight function of non-local mean filtering. Based on the theory mentioned above, this paper proposes a non-local mean filtering method for suppressing abnormal noise based on the similarity of seismic waves to effectively suppress random noise and abnormal amplitude noise in seismic signals. The trial results of theoretical and actual seismic data show that the method proposed in this paper can effectively suppress random noise and abnormal amplitude noise in seismic signals, and can further improve the signal-to-noise ratio of seismic data.