Abstract: In order to increase the oil reserve and production, use of effective methods for reservoir prediction has become the main requirement of oil exploration. Owing to the limitation of resolution, reservoir prediction based on conventional seismic profile is hard to meet the needs of oil field production. This paper devotes itself to the deep study on the wavelet decomposition and reconstruction theory. According the actual data, a reservoir geological model which can reflect the actual situation is created. Forward modeling is done based on the model and the seismic characteristics of specific geological objects are revealed. It is found that the low frequency part of seismic data contains some useful information on the spatial distribution of reservoir. Model tests and practical seismic data interpretation have demonstrated that the method can fully dig out valid information from seismic data and effectively predict the distribution pattern of the reservoir. The result of reservoir prediction using the method coincides well with drilling results.