The invention discloses a seismic data noise suppression method based on a block matching algorithm and singular value decompression. With the method, problems that noises are not suppressed completely and effective information is attenuated because inherent local and non-local similarity of events of the seismic data is not considered can be solved. The method comprises the four steps of similarity grouping, three-dimensional transformation singular-value filtering and denoising, re grouping based on basic estimation and wiener contraction coefficient calculation, and wiener filtering. According to the method, in theory, a matching algorithm is used; with consideration of the inherent characteristics of the local and non-local similarity of events of the seismic data and combination of two-dimensional singular value decompression and one-dimensional singular value decompression, sparse representation of seismic data is realized by using structural information of the seismic data fully. Therefore, effective signal attenuation is reduced to the great extent and the noise suppression effect is improved.