The study investigated the empirical relationships among social power, message behavior, and compliance. A conceptualization of social power from power dependency theory was offered, and two models of the general relations among these constructs were advanced. They included a mediation model and a moderating model. These issues were investigated by having 108 subjects participate in a bargaining experiment where power was controlled by systematically varying each partner's alternatives, and participants' message behavior and outcomes were observed. Although power had non-trivial effects on message behavior, the data were generally inconsistent with the mediation model. The data were consistent with the moderator model which predicted that message effectiveness varies as a function of social power. Implications and limitations of the results are discussed.