Abstract: The identification and prediction of coalbed methane reservoir are an important problems. The conventional inversion method is not very accurate in reservoir prediction and hard to satisfy the actual needs. The nonlinear stochastic inversion method is used to improve the accuracy of reservoir prediction by taking drilling and logging data as the constrained conditions and by extracting the information of amplitude, frequency and phase of the target stratum seismic wave. Compared with the actual geological data, the method has been well applied in Yushe-Wuxiang CMB block, Qinshui coalfield. The consistent rate of prediction precision and drilling-revealed results is high, and the predicted distribution of CMB reservoir thickness coincides with geological law.