Since the traditional conflict parameter can’t measure the degree of similarity between the evidence effectively, a combination method of evidence based on classification correction is proposed to solve the high conflict paradox and “0” paradox. Three parameters are used to measure the degree of similarity between the evidence together, including the evidence distance parameter, the conflict parameter and the direction parameter. According to the three parameters, the evidence is classified into four categories, including the consistent evidence, the no-conflict evidence, the low conflict evidence and the high conflict evidence. The different evidence is given different correction coefficients, and the revised evidence is combined by means of Dempster rule. Numerical experiments show that the proposed method can solve the high conflict paradox and “0” paradox well, and keep the good mathematical properties of evidence theory.