Although policing has developed a pivotal role in helping individuals that are going through mental crises (Livingston, 2016), very few studies have sought to explain how officers assess one's cognitive competences during police encounters. I propose that the procedural infrastructure for interaction provides an environment in which a gradient of accountability enables inferences about participants’ orientations to mental health-related phenomena. This research thus aims to describe the interactional mechanisms through which police officers categorize individuals as mentally ill (explicitly or implicitly), as well as the consequences of this social category for the ongoing interaction. Using a conversation analytic approach, I will analyze over 60 dashcam videos obtained from the Seattle Police Department as part of a larger project entitled “Talking Justice: Identifying Interactional Mechanisms to Improve the Quality of Police-Civilian Encounters.” Preliminary findings suggest that orientations to a “mentally ill” social category are evident when civilians systematically fail to provide police officers with sequentially coherent responses, thus disrupting what Garfinkel (1967) calls natural facts of life. Addressing issues of mental health as concerns faced in situ by police officers and members of the public allows us to elaborate on the social embeddedness of a phenomenon often treated as phenotypical (Maynard, 2019), which could translate into better informed mental health training programs for police officers.