The data consist of ratings in several dimensions of interactions as they occurred in police-citizen encounters systematically observed in randomly selected police patrols in St. Paul, Minnesota. The theoretical model proposed views social interaction as a process which unfolds over time as two or more participants, in the context of some task, choose and carry out actions which have consequences for both participants. The focus of the study is on the response of each participant in the encounter to the actions of the other participant. Fundamental elements of this social interaction include the contingency of present events on past events, the character of the task, the underlying decision processes, the nature of the participants, and the interaction sequence which results. A series of Markov models is examined with the use of log linear models. A second-order Markov model with heterogeneous roles was found to provide the best fit for the data. The dynamic character of the model is explored mathematically by examining its eigenvalues and eigenvectors. The results support the view of an interaction over time of participants enacting complementary roles, with citizens being compliant and submissive and officers being directive and controlling. Implications of these findings for the enforcement process are examined. It is suggested that future research explore social interaction in a variety of settings and further examine possible decision rules which may account for the observed behaviors. Appended are issues arising from the use of mathematical models and a test for mean recurrence times. Tabular data and 180 references are provided. (Author abstract modified)