On the basis of the well‐established association between early alliance and outcome, this exploratory study investigated the associations between the therapeutic alliance and long‐term outcome, 3 years after treatment termination. In addition to the early alliance, pre‐treatment patient characteristics and expectancies that were significantly related to early alliance were included in the statistical analyses. The data are from the First Experimental Study of Transference, a dismantling randomized clinical trial with long‐term follow‐up. One hundred out‐patients who sought psychotherapy due to depression, anxiety and personality disorders were treated. Alliance was measured with Working Alliance Inventory after session 7. Change was determined using linear mixed model analyses. The alliance alone had a significant impact on long‐term outcome of the predetermined primary outcome variables of the study: Psychodynamic Functioning Scales and Inventory of Interpersonal Problems. Contrary to common clinical wisdom, when the pre‐treatment patient variables were included, more personality disorder pathology was the strongest predictor of favourable outcome, over and above the effect of the alliance, which was no longer significant. Clinical implications are discussed. Copyright © 2012 John Wiley & Sons, Ltd. Key Practitioner Message Patients with more personality pathology responded more favourably to long‐term relational dynamic psychotherapy than patients with less personality pathology.Patient characteristic had stronger effect on long‐term outcome, over and above the effect of alliance.In relational dynamic psychotherapy, it is important to focus on dysfunctional interpersonal problems, as they appear both in the transference and in the patients' day to day life outside therapy.The therapeutic task is to balance a forthright discussion of recurrent patterns of interpersonal difficulty with an empathic appreciation that these patterns are hard to change because of their longstanding entrenched nature.