This paper focused on the low-frequency shadow caused by the attenuation by absorption of seismic waves in viscous media, which can be used as a hydrocarbon indicator. It is very difficult to detect the low-frequency shadow in deep carbonate reservoir by conventional time-frequency analysis methods, due to the poor seismic resolution resulted by the complex inner structure in stratigraphic and lithologic traps. A new matching pursuit algorithm based low-frequency shadow method for carbonate reservoir characterization was introduced. This paper described the low frequency shadow and the principle of matching pursuit algorithm based on spectral decomposition of seismic signal, and showed that the method has high time-frequency resolution compared with others and it can be used to predict the deep carbonate reservoir. The well drilling data proved the good prediction efforts.