Data fusion technology is widely used in automatic target recognition systems to improve the efficiency. In the framework of evidence, information fusion relies on the use of a combination rule allowing the belief function to be combined. However, Dempster’s rule of combination is a poor solution for the management of the conflict between the various information sources. It is proved that, if the false evidence can be correctly selected, Dempster’s rule can deal with highly conflicting evidence combination efficiently. However, how to determine the false evidence is an open issue. In this paper, a novel method to determine the discounting coefficient is proposed. First, the distance function between bodies of evidence is introduced to express the degree of conflict degree. Then, the confidence lever of each piece of evidence is obtained to reflect the reliability of each information source to some degree. The discounting coefficient can be determined finally through the relative credibility of sensor reports. The numerical example of multi-sensor fusion target recognition based on DS theory is shown to illustrate the efficiency of the presented approach.