Abstract: In the prediction of CBM reservoir parameters, the redundancy of seismic attribute data brings inconvenience to the prediction. In order to obtain the optimal reduction of seismic attributes of CBM reservoir, an method using improved cross entropy algorithm to optimize seismic attributes was proposed. This method improved the samples generated by the iterative process, and then generated an excellent set of samples. At the same time, the model of rough set attribute reduction was used as the objective function to solve the optimization. The model of attribute reduction of rough set is solved by using the improved combined cross entropy algorithm. Finally, the seismic attributes in an area were used to verify and compare the algorithm with the other algorithms. The results show that the algorithm is less time-consuming, has high precision and ratio of reduction, and can be effectively used in seismic attribute reduction of CBM reservoir.