This paper describes a Chinese continuous speech recognition system in which we use a new statistical model, i.e. the State Transition Model(STM) to express phoneme and word in Chinese. The acoustic-phonetic process of this system can be divided into two levels. The first level is phoneme level process which is carried out before linguistic process and provides the fundamental information of the linguistic process. The second level is word level process which is carried out after the linguistic process and provides the information of coarticulation of the consonant and the vowel in a word. For linguistic process, 12 kinds of Chinese sentence patterns most in use, have been registered in the form of the context-free grammars. Semantic analysis is carried out at the same time with the syntactic analysis. Experiments of phoneme, word and sentence recognition have been done to evaluate the performance of this system.