Despite the considerable research on mediator behavior, the cognitive structures and processes that presumably guide the strategic and tactical choices of professional mediators are poorly understood. The current study made use of a reflective case study method to explore in considerable detail the strategic thinking of five experienced mediators. The project was conducted at the National Institutes of Health whose Office of the Ombudsman (OO) mediates disputes among the institute’s scientists. Eighteen cases were studied. The thinking of the mediators in these cases displayed regularities that are described in terms of the ombuds team’s working mental model of mediation. The mental model consists of two strongly contrasting intervention scripts: a deep problem-solving script (DPS) focused on identifying and addressing latent issues of an interpersonal or systemic kind and a tactical problem-solving script (TPS) focusing instead on the issues as presented by the parties. The tactical script was applied in either an integrative bargaining mode or a more distributive quasi-arbitration approach. The choice of which script to follow in a given case is determined by first order decision rules concerning the existence and nature of any latent problems that may be present, and second order decision rules concerning the parties’ capacity to engage in “deep” problem-solving. Despite their very different foci, both DPS and TPS appear to follow the same metascript of problem-solving stages, beginning with an intensive diagnostic phase during which the decision rules are applied and a script “selection” is made. DPS is the preferred intervention mode of team members. Every case began with at least a preliminary effort to search for and address latent causes, and team members expressed dissatisfaction if they could not apply DPS in cases where latent problems were thought to be fueling the conflict. However, ombudsmen used the scripts flexibly and switched to TPS if DPS was unnecessary or impractical. Both scripts produced agreements that were useful to the parties and to the institution’s scientific purposes, particularly the fostering of scientific competence. The mental model is heavily shaped by the social context in which the ombudsmen function. Thus, the primacy of DPS in the model appears to be due to the fact that the ombudsmen are “repeat players” in the life of the NIH and therefore in a position to become adept at recognizing the latent sources of its dysfunctional conflicts, are under a strong role mandate as ombudsmen to pay attention to covert patterns of organizational dysfunction, and deal with disputants pressed to address latent issues blocking scientific work.