The construction of a network information text categorization model is proposed. Accroding to the feature of network datagram, the keywords are selected as classficatoin features and the keywords frequencies are counted to construct a text categorization model. The experiments show that the K-nearest neighbour method is superior to the nearest neighbour method.