According to the morphology and sparse signal theory, morphological component analysis (MCA) method is used for random noise attenuation in seismic data. The key of MCA is to select the appropriate dictionaries. In view of the characteristics of seismic data and computational complexity, UWT dictionary and Curvelet dictionary are selected. BCR algorithm is used to solve objective function, and the denoised results are obtained by decomposing the seismic data into two morphologically different components and discarding the random noise which can’t be sparsely represented in dictionaries efficiently. Theoretical and real data processing verified the efficiency of MCA method.