For the problem of not fully using information and how to optimizing the weight of buffer operator with variable eight, a new buffer operator with smooth variable weight is proposed. The property of the new buffer operator is studied, nd it is proved that the operator has weakening effect to the sequence, furthermore, operator’s recursive inequation of adjust  ntensity is obtained. Then a multi-criteria optimization method is proposed for optimizing weight, object functions are onstructed and optimal weight is determined by using the genetic algorithm. Finally, the example analysis shows that the uffer operator with smooth variable weight can improve the fitting accuracy of grey prediction model effectively.