This article explores the relationship between dialogue and power in the practice of dialogue-oriented, "not-knowing" forms of therapy. It is argued that power of a dynamic and reversible kind infuses much ordinary social dialogue, and that the joint processes of power and resistance work together to render an interaction dialogical. In contrast, in dialogical therapy, overt exercises of power threaten the interaction's dialogical status, and power is deferred and denied by the therapist through not-knowing practices. A case study of Harlene Anderson's (1997) is used to illustrate that it is precisely power's presence that informs the practices of not knowing and uncertainty that characterize dialogical therapies. It is suggested that the not-knowing therapist withholds aspects of his or her voice as a condition for dialogicity. Instead, special speaking arrangements are required, in which the therapist's not-knowing is continuously communicated to the client, for the therapeutic conversation to remain dialogical. Without these speaking arrangements, I argue that therapy moves toward monologue. Therapists inherit powerful speaking positions from the institutional and sociocultural context, and the rejection of power within therapy serves only to conceal this aspect of power, which nevertheless pervades the therapeutic relationship. Finally, it is suggested that power is a "common factor"--shared by all therapies--and that our status as "inheritors" of power needs to be included in our understanding of the therapeutic process. This expanded view of therapy requires the re-theorization of dialogue, such that it includes, rather than excludes, considerations of power.