There are many methods of deconvolution at present, based on conventional convolution model. In conventional convolution model, supposing the wavelet not to be attenuated, so it doesn't accord with actuality that the amplitude attenuation gradually. Therefore, it's necessary to establish a nonstationary convolution model which can describe the nonstationary wavelet. Based on the convolution model, nonstationary deconlution can improve the resolution of seismic data without compensatting amplitude of seismic trace. Comparing the results of spiking deconvolution and nonstationary deconvlution in the paper, the analysis of theoretical model and real data computation prove this method is right.