Abstract: In this paper, on the basis of research on the application of variable precision rough set model in building decision tree, a method of building decision tree (DT) with rules that have definite confidence is proposed. It is an improvement to the method of building DT, decision tree built in this way is more practical and comprehensible. It also gives out the solution to the problem of how to choose the better attribute as the node when two or more attributes have the same value of quality of classification. Compared to the traditional ID3 algorithm, decision tree built with this method is not only simple-structured, but also more practical and comprehensible.