Although the therapeutic alliance is a consistent predictor of psychotherapy outcomes, research has not distinguished between the roles of patient and therapist variability in the alliance. Multilevel models were used to explore the relative importance of patient and therapist variability in the alliance as they relate to outcome among 331 patients seen by 80 therapists (therapist average caseload was 4.1). Patients rated both the alliance and outcome and all models adjusted for baseline psychological functioning. The results indicated that therapist variability in the alliance predicted outcome, whereas patient variability in the alliance was unrelated to outcome. Reasons why therapist variability as opposed to patient variability predicted outcome are discussed. Clinical implications include therapists monitoring their contribution to the alliance, clinics providing feedback to therapists about their alliances, and therapists receiving training to develop and maintain strong alliances.