To try to decrease the preference of the attribute values for information gain and information gain ratio, in the paper, the authors puts forward a improved algorithm of C4.5 decision tree on the selection classification attribute. The basic thought of the algorithm is as follows: Firstly, computing the information gain of selection classification attribute, and then get an attribute of the information gain which is higher than the average level; Secondly, computing separately the arithmetic average value of the information gain ratio and information gain of the attribute, and then select the biggest attribute of the average value and set up a branch decision; Finally, to use recursive method to build a decision tree. The experiment shows that this method is applicable and effective.