Two problems in previous research on the demeanor-arrest link raise questions about reports that hostility directly increases the likelihood of arrest. First, demeanor frequently is defined so broadly as to include criminal acts. Second, criminal conduct is not controlled satisfactorily. Because extant reports of hostility effects may be misleading, more rigorous tests of the hostility thesis are need. The data used in this paper were drawn from an observational study of police behavior conducted in Dade County, Fla., during 1985 and 1986. Trained civilian observers accompanied randomly selected officers from three districts of the Metro-Dade Police Department on 877 8-hour patrol shifts. The observers recorded selected aspects of traffic stops, crimes in progress, and interpersonal disputes in observation schedules designed specifically for each type of encounter and wrote brief narrative reports that summarized each interaction. During the course of observation, study officers handled 245 disputes in which they contacted at least two involved parties on opposite sides of a conflict and in which observers recorded information about each of several variables of interest. In an analysis that used a demeanor measure that does not confound crime and that controls for crime more comprehensively, this study found that displays of hostility that violate no laws do not increase the likelihood of arrest in and of themselves. The implications of this finding are discussed. 4 tables and 33 references