Abstract: Multiples attenuation is critical for single channel seismic data processing, and the adaptive matched filtering is one of the key techniques to be used for the purpose. However, the traditional matched filtering algorithms are not always adequate to the description of nonstationary seismic signal changes with time and space. To solve this problem, the regularized nonstationary regression is proposed. This paper studies the application of regularized nonstationary regression, starting from basic principles, then testing regularization parameters using model data to pursue their influence on processing effect, and finally applying this technology to predict subtraction process of field data. As the result, multiples are effectively attenuated. Practical application demonstrates that the method has great advantages comparing to the traditional method in multiple attenuation.