Most psychotherapy research uses a one-with-many design, in which each therapist (the one) treats multiple clients (the many), which raises the challenge of nonindependent data. We present a statistical model for analyzing data from studies that use a one-with-many design. This model addresses the problems associated with nonindependence and can address theoretically relevant questions. To illustrate this model, we analyzed data in which 65 therapists and their 227 clients rated their therapeutic alliance. The primary finding was that both therapist and client alliance ratings were largely relational (i.e., specific to the unique therapist-client combination). There was little consensus among clients treated by the same therapist about the quality of the therapeutic alliance, although some therapists reported forming stronger alliances than other therapists. There was substantial dyadic reciprocity, indicating that if a therapist reported an especially good alliance with a particular client (better than with his or her other clients), then that client was also likely to report an especially good alliance (better than those reported by the therapist's other clients). (Contains 1 figure, 4 tables and 4 footnotes.)