Three models describing the structure of talk and silence sequences within and across conversations presented in a previous report (Cappella, 1979) are tested. The Markov model, describing talk and silence sequences within a conversation, is found to be a valid representation on a dyad‐by‐dyad basis. The Independent Decision (ID) model shows some predictive validity between conversations, although its “fit” within the conversation is less than the Markov model. The Incremental model in relaxing the consistency‐across‐conversation assumption of the ID model finds differences due to switching of partners in the probability of breaking or continuing mutual silences and in the probability of continuing to hold the floor. The implication for deriving dyadic interaction patterns from individual interaction styles a