Abstract：Evidential network is a reasoning model extend from Demspter-Shafer(DS) theory. Unite-believe value(UBV) is a way of deducing in evidential network. In a general way, the parameters in the deduction are given by expert knowledge. A method of getting optimized parameters automatically is presented. In which the value of conflict is used as adjudge function and the genetic algorithm based on gray code is used. The simulation experiment shows that the result of the proposed method is more accurate than the result of the method using the parameters given by expert knowledge in emulation, and the more population at beginning, the better accuracy of the recognition result.