Abstract: It is first analyzed how much information is used when recognizing a Chinese character. It is indicated that the single character recognition algorithm is an equal probability model and needs the most information. So the Chinese text is regarded as a Markov model, which means that the character is determined by the last m characters. On the basis of the statistics of the text, a lot of Chinese linguistic knowledge is obtained. An automatic recognition is designed, in which the character is matched in the next neighboring character set of the last character. After recognized, the sentence is treated as the linguistic knowledge before it is output. So the recognition speed and recognition rate are higher than that of the single character recognition algorithm.