In this paper, I proposed a fast dictionary-learning based seismic denoising approach using the sequential generalized K-means (SGK) algorithm. In the SGK algorithm, each atom in the dictionary is updated by an average of several sample signals while K-SVD uses computationally expensive SVD to update each atom. Thus, the SGK algorithm can be much faster than K-SVD algorithm for adaptively learning the dictionary. I applied both K-SVD and SGK to dictionary learning of seismic data for random noise attenuation. The results from three different examples show that SGK is much faster than K-SVD without sacrificing much denoising performance. I suggest substituting the K-SVD with SGK in any applications that require sparse coding. Future research direction may include applying the SGK based dictionary learning for multidimensional seismic data reconstruction.