Sandstone is close to mudstone in P-wave velocity，density，P-wave impedance and other conventional parameters sometimes. So，it is difficult to solve the reservoir prediction problems and to discriminate gas-bearing sandstone from shale by using general reservoir prediction methods such as post-stack impedance inversion and single parameter prediction. Because of multiple solutions of seismic data，the result of reservoir prediction is always at risk of failure. Therefore，based on seismic rock physics analysis，combined with multi-parameter probability fusion method to fuse petrophysical parameters and geophysical parameters，pre-stack reservoir prediction was carried out，which may reduce multiple solutions and enhance prediction precision with fewer risks. It is feasible，and also reveal and characterize the uncertainties in reservoir prediction. The results of the study show that pre-stack seismic inversion parameters such as P-wave velocity and impedance，S-wave velocity and impedance，longitudinal wave velocity ratio and Poisson’s ratio may discriminate sandstone from shale and identify gas-bearing sandstone to some degree，but the precision of prediction results were not high. The method of multi-parameter probability fusion may extract as much as possible information of reservoir for prediction. Therefore，the method not only can effectively discriminate the lithology and distinguish gas-bearing formation from non-gasbearing formation，but also improve the precision of reservoir prediction. At the same time，the precision of the thickness of sand layer and gas layer predicted by multi-parameter probability fusion method is higher than that of conventional elastic inversion method.