Abstract: Presents the new method using Hidden Markov Models（HMM） to supervise document classification.Represents the document to be classified in a kind of hidden Markov models.The states of HMM eject the symbols with a certain probability.These symbols composes of the classified documents.The class of document is supposed to be composed by some character item series.By calculating the output probability of the HMM on the class character series can get the max corresponding output probability and the output series.Compares the result on all the class can decide the category of a certain document.The model is evaluated on the real dataset with Naive Bayes,KNN and simple vector models.It is shown to be successful method in text categorization.