In the process of constructing decision trees, the selecting criteria of classification attributes will directly affect the classification results. Here we presented the classification contribution function (CCF), a new concept based on rough sets theory, which is regarded as the criteria for choosing attributes in the core of attributes. The basic idea of CCF is using of discernibility matrix to determine attribute core (if there is no core, using its reduction). Then through the classification contribution function to determine core classification contribution value and employ the big value as node. Next employing the selected attribute ways to divide decision-making system and each value attribute can produce a subset. The experiments show that, being compared with the entropy based C4.5 and weighted mean roughness, our method can get simpler decision tree and improve the efficiency of classification.