Abstract: During the acquisition of vertical seismic profile (VSP) data, coupling noise between casing strings frequently emerges due to poor well cementing or the presence of cement annulus between multiple casing layers. This type of noise features high energy, a wide spatial influence range, and slow attenuation. Furthermore, it overlaps effective signals in multiple domains, such as the time and frequency domains. Therefore, casing coupling noise severely affects the signal-to-noise ratio of VSP data. This study first introduces the basic theory of Morphological Component Analysis (MCA). Then, it conducts a comparative analysis of the distribution of effective signals and casing coupling noise in the time, frequency, and time-frequency domains. Subsequently, this study processes the VSP data as a combination of effective signals, casing coupling noise, and random noise through modeling. Finally, it proposes a MCA-based method for suppressing casing coupling noise mixed in the VSP data. In this method, corresponding overcomplete dictionaries are constructed utilizing the tunable Q-factor wavelet transform (a sparse representation dictionary for effective signals) and the cosine transform (a sparse representation dictionary for the casing coupling noise). Then, the block coordinate relaxation method is employed to suppress the casing coupling noise in the VSP data. As indicated by the processing results of actual VSP data, the method proposed in this study enjoys high adaptability to the changes in the energy and frequency-domain distribution zones of casing coupling noise. Furthermore, it exhibits high fidelity to effective signals while significantly suppressing such noise.