Video fragments allow for intensive, close-up analyses of what actually happens in violent incidents because the image can be paused, and repeated again in slower or faster paces. In this project, we ask how bodily behaviour of group members, including bystanders, is related to the escalation and de-escalation of antagonistic situations. We use video footage of violent and near-violent situations in public spaces, found on the internet. The coding scheme we developed aims to capture how participants build up the tension and how they try to dampen it in these situations. For instance, we code participants’ aggressing behaviour (threats or attempts to invade the body space of the opponent), the ways they move around, and how they try to de-escalate. On the basis of our coding, we are going to analyse whether it makes a difference if group members are aligned (engaging in the same behaviours) and whether that matters for being successful in launching or fending off an attack. Furthermore, we will analyse whether incidents in which a lot of aggressing occurs end more violently. Also, we are going to relate the occurrence and the duration of bodily behaviours to (unequal) group sizes: does de-escalation appear more often when more bystanders are present, and does it matter to which group these bystanders belong?