Abstract: A feature based Chinese part of speech tagging model is presented, which is an extension of hidden Markov model This model not only takes the contribution of hidden state at time t into account, but also considers the contribution of the observation at time t+1 Since the number of observation is big, it is not practical to use the observation state matrix So in this paper each word is tagged with features, and then the matrix of feature state matrix is trained, which is much smaller The balance of accuracy and practicality is achieved