The therapeutic alliance has historically emerged as a pantheoretical correlate of favorable psychotherapy outcomes. However, uncertainty remains about the direction of the alliance-outcome link, and whether it is affected by other contextual variables. The present study explored (a) if early alliance quality predicted subsequent symptom change while controlling for the effect of prior symptom change in interpersonal psychotherapy (IPT) for depression, and (b) whether baseline patient characteristics moderated the alliance-outcome relation (to help specify conditions under which alliance predicts change). Data derived from an open trial of 16 sessions of individual IPT delivered naturalistically to adult outpatients (N = 119) meeting criteria for major depression. Patients rated their sociodemographic, clinical, and interpersonal characteristics at baseline, their alliance with their therapist at Session 3, and their depressive symptoms at baseline, after every session, and at posttreatment. Data were analyzed using hierarchical linear modeling. Results indicated that alliance quality did not predict subsequent depression change, controlling for prior depression change. However, a significant education by alliance interaction emerged in predicting quadratic depression change (γ = .0007, p = .03); patients with higher levels of education who reported good early alliances with their therapists had the most positively accelerated change trajectory (i.e., faster depression reduction), whereas patients with higher levels of education who reported poorer early alliances had the most negatively accelerated change trajectory (i.e., slower depression reduction). The findings may help clarify a specific condition under which alliance quality influences subsequent improvement in an evidence-based treatment for depression. (PsycINFO Database Record