Abstract: The singular spectrum analysis method has been widely applied to seismic data denoising. However, when applied to seismic data with developed faults and fractures, the fault information and noise components tend to overlap in the singular value spectrum. This overlap may result in the blurring of fault features, especially the fault details, during the process of noise suppression. To address this issue, a fault-preserving singular value spectrum analysis method was developed. Firstly, the generalized S-transform was introduced to enhance the method’s adaptability to non-stationary signals. Secondly, comprehensive fault attributes were extracted using the singular value spectrum analysis method, based on which a filtering constraint weighting function was constructed. This function was then incorporated into the singular value spectrum filtering process to enable adaptive adjustment of filter weights at discontinuities, thereby avoiding damage to faults and other discontinuous structural features. Finally, a singular value attenuation function was designed to perform attenuation-based filtering on the singular value, achieving effective noise suppression while preserving fault characteristics. Both synthetic model tests and real data applications indicate that the proposed method can efficiently suppress noise while better preserving fault and other discontinuous information.