Objective: While previous research agrees that third-parties often manage to de-escalate interpersonal conflicts when they intervene, we still know little about how they achieve this influence. The aim of the present study is to address this gap in the literature by investigating how third-parties de-escalate conflicts. Method: We conduct a two-part qualitative analysis of closed-circuit television (CCTV) footage of 48 reallife conflicts from the streets of Amsterdam. The first part consists of an inductive analysis of CCTV-footage investigating the typical sequence of de-escalatory interventions. The second part consists of a deductive coding based on the findings from part one of the analysis. Results: We identify an ideal-typical model of de-escalation consisting of three phases: Objection, separation, and placation. This model describes how third-parties adapt their intervention to the reaction of the antagonists of the conflict through a contingency principle: When the current phase of intervention fails to de-escalate the conflict, the third-party proceeds to the following phase of the model. We also identify observable intervention behaviors that are characteristic of each of the three phases. Conclusions: The findings demonstrate that there is not one way to de-escalate a conflict, but rather that third-parties are successful, because they adapt their intervention to the situation at hand. The findings of this study imply that if we want to get closer to understanding when third-party interventions are effective, we need to acknowledge the complexity of these interventions and move beyond the action/inaction dichotomy.