The combination of conflicting evidences has been one of the important issues in Dempster-Shafer theory since the example of Zadeh paradox produced by Dempsters rule of combination was found. There is no solution accepted universally as yet. In this paper, an approach to handling this problem is proposed. Before adopting the Dempsters rule to combine belief functions, the assignments of mass functions are pretreated by the disturbance of ignorance. The pretreatment is in favor of the normalization of combination. Compared with other alternatives, the new approach is consistent with Dempsters rule in the form and flexible to combination, especially, it can be theoretically explained by the generalized Bayesian formula.