Abstract: A new method of combination forecast is proposed in this paper. Determining the weighting coefficient in a combination forecast model is translated into estimating significance of the attributes among rough set. Knowledge entropy is computed and a computation approach of weighting coefficient in a combination forecast model is showed on the basis of establishing a relation data model about combination forecast. The proposed approach overcomes the subjectivity of traditional methods, avoids solving linear of nonlinear extreme problem, automatically collects predicting model, is simple in calculation and makes combination forecast more objective. The validity of the proposed approach is verified with a case.