Abstract: A novel method on network learning is proposed. It can optimize the knowledge map in the global conditions and the necessary and sufficient requirements. All the knowledge units and their relations, stored in a generalized list, are viewed as the vertices and edges in the knowledge map. After testing the knowledge units which have been learned by users before, we can get the knowledge levels corresponding with each user. The sort sets of knowledge units can be deduced according to the algorithms of knowledge unit central and degree of difficulty, and then the complete sets of knowledge map paths are achieved. The approximation is substituted by the precision in the proposed method, and the defect that the necessary and sufficient conditions cannot be satisfied simultaneously is solved as well. The network navigation paths in our knowledge map can be used to avoid the problems of disorientation and information overloading, which can improve the efficiency of user learning on the network.