Abstract: The methods of cbranch classification adopted universally at present can only provide a locally optimal solution. Firstly, a theorem equivalent to the principle of Parsimony is put forward in this thesis. And then, some possible cases are discussed when the method of maximal synchronization is used for classification and a few relevant theorems are given. On this basis a example counter for the method of maximal synchronization is raised and an amendment algorithm is given. The experimental results show that thes algorithm is correct and feasible and at least as good as the method of maximal sychronization.