Abstract: Decision tree is one of heated fields in data mining，and it is a widely-used solution for classification problems.But the design of the optimal decision tree has been proved to be NP-hard.This paper first introduces the main thoughts of algorithm of ID3 ，then imports the conception of general correlation function in order to make up the weakness，and puts forward an algorithm of structuring decision trees.General correlation function between conditional attributes and a decisive attribute is the criteria of attribute selection in the algorithm.What’s more，a contrast to ID3 is made by experiments.Results demonstrate this algorithm not only optimizes decision trees model，but also improves classification accuracy.