In the framework of evidence theory, data fusion is to build a single belief function by combining several belief functions derived from distinct information sources. Dempster's rule of combination, a classical combination rule with several interesting mathematical properties, is widely employed. However, as an inherent problem, Dempster's rule of combination is incapable of managing the existing conflicts from various information sources at the step of normalization. The conflict management becomes an important problem in the operation of combination, especially in the condition of highly conflicting, which leads to produce the illogical result of combination. In this paper, we introduce the idea of grey relational analysis (GRA), and propose a new conflict reassignment approach of belief functions, as a preprocessing method, to automatically identify and reassign the conflicts on belief functions before combination. Three numerical examples are employed and implemented. Through analysis and comparison of the combined results, of the existing alternatives, the proposed approach not only can flexibly solve the problem of conflict management with better convergence performance, but also can automatically evaluate the reliability of the information sources and distinguish the unreliable information.