Objective: Evidence is inconclusive on whether variability in alliance ratings within or between therapists is a better predictor of treatment outcome. The objective of the present study was to explore between and within patient and therapist variability in alliance ratings, reciprocity among them, and their significance for treatment outcome. Method: A large primary care psychotherapy sample was used. Patient and therapist ratings of the working alliance at session three and patient ratings of psychological distress pre-post were used for analyses. A one-with-many analytical design was used in order to address problems associated with nonindependence. Results: Within-therapist variation in alliance ratings accounted for larger shares of the total variance than between-therapist variation in both therapist and patient ratings. Associations between averaged patient and therapist ratings of the alliance for the individual therapists and their average treatment outcome were weak but the associations between specific alliance ratings and treatment outcome within therapies were strong. Conclusions: The results indicated a substantial dyadic reciprocity in alliance ratings. Within-therapist variation in alliance was a better predictor of treatment outcome than between-therapist variation in alliance ratings.