This paper deals with a newly proposed identification problem of the turn-taking (surface level exchange) pattern of spontaneous conversation using only the power patterns of two conversants. Some necessary concepts of turn-taking are first defined corresponding to the power pattern. A new algorithm, HSPN (Hidden Stochastic Petri Net) is introduced to identify Conversation Unit (CU), that is, turn-taking patterns. HSPN is equivalent to a multi channel coupled Hidden Markov Model (HMM), which incorporates synchronization of inter channel states. It is transformed into an ordinary Markov model using HSPN. A high CU correct rate is obtained for spontaneous conversation data.