Drawing on models of the divorce adjustment process, conflict theory, and previous research, a number of hypotheses were derived about the influence of divorce process variables on disputant behavior in mediation and mediation outcome. Premediation levels of attachment, anger, and sadness were assessed in 112 couples and the mediation sessions offifty of these were videotaped. An existing coding instrument was modijed to quantify disputant behaviors. Specijc behaviors in mediation were found to be predictive of mediation outcome. Antecedent attachment and anger were found to be predictive of behavior in mediation and mediation outcome. In this article, the impticationsfor mediation practice are discussed.Empirical research in divorce mediation is limited in quantity and scope. The majority of efforts have been directed toward demonstrating efficacy, as the field has needed to establish its credibility and jostle for position within an already crowded marketplace. Although generally positive, the efficacy research has demonstrated considerable variability in outcomes (Kressel, 1985), suggesting that the types of intervention and the various models of divorce mediation differ in their degree of effectiveness. However, much of the research has treated the mediation process as a "black box" (Benjamin and Irving, 1995), comparing intake data and agreement rates with limited reference to what actually transpires in mediation. Consequently, there is little common understanding of the mechanisms involved and the circumstances under which mediation is more or less effective (Kressel, Butler-DeFreitas, Forlema, and Wilcox, 1989).Part of the difficulty is the absence of a coherent theory to explain how individuals in conflict behave in mediation and how mediators can intervene