Traditional methods of documents classification need characteristic abstraction and classifier training. The work of collecting trainable text terms is laborious and time-consuming. Additionally, it is difficult to abstract the characteristics from Chinese documents. In order to solve the problem, this paper proposes an ontology-based approach to improve the efficiency and effectiveness of web documents classification and retrieval. Firstly, the approach establishes an ontology model based on Hownet[6] kownledge base and its method. Then, it creates ontologies for each subclass of the classification system. It uses RDFS to convert Hownet into ontology and to define the relations among ontologies. The web documents classification is performed automatically using the ontology relevance calculating algorithm. Comparing with the method of KNN[2], the results of our experiments indicate that the accuracy of ontology-based approach is close to KNN, its algorithms is more robust than KNN, and its recalling rate is better than KNN.