Mediation is one of the most widely used methods of conflict resolution, yet causal evidence on when and why it succeeds remains limited. This paper presents experimental evidence on the conditions under which a neutral third-party mediator can enhance the efficiency and fairness of two-person bargaining outcomes. In a controlled laboratory setting, participants perform a real-effort task in which it is ambiguous whether performance differences reflect effort or luck, with relative performance determining their outside options. We compare a baseline bargaining treatment to three mediation interventions: bargaining with a mediator’s recommendation (T-MED), bargaining informed by the mediator’s perception of fairness (T-FAIR), and bargaining with an aggregate recommendation drawn from multiple mediators (T-NORM). Contrary to common wisdom, mediation does not systematically increase agreement rates, shorten negotiation time, or improve outcome fairness across treatments. Why is this the case? Our preliminary results suggest that mediation is effective only when stakeholders trust the mediator: Stakeholders who trust the mediator reach more agreements, negotiate more quickly, and achieve fairer outcomes than in the baseline, especially under the aggregated norm treatment (T-NORM). Moreover, aggregated recommendations tend to enhance fairness when they are close to the midpoint of stakeholders’ perceived entitlements or when conflict (measured by stakeholders’ self-serving bias) is low. These findings highlight the central role of trust in determining when mediation is (or is not) effective and underscore the potential of social norms to foster more efficient and fair mediated outcomes.