The web documents classification is an important research content of web mining. Document vector space model is the foundation of automatic classification of documents, while it is difficult to eliminate redundant attributes and reduce the dimension of the vector space. The Rough Sets Theory is applied to generalize the information system comprised by document samples set, and to compute the best reducing properties set. So dimension of document feature space is reduced greatly, and disturbance to document classification is decreased too, which improve the efficiency of classification. In addition, using the advantage of adaptive classification and incremental learning of Fuzzy ARTMAP neural network, the online adaptive classification of web document is achieved.