Background A good therapy relationship is viewed by many as a precondition for effective therapeutic work, and by some even a therapeutic factor in itself. In theory, the therapy relationship has been divided into the components working alliance, transference, and “real relationship”. Most empirical research has been conducted on the working alliance, which concerns the quality of collaboration between therapist and patient. Methods State-of-the-art position statement. Results Research on the working alliance has gone from fairly simple pre-post correlational designs to multi-level cross-lagged panel designs. Although almost all research is based on observational data, the “new generation” of alliance research takes account of temporal precedence and potential unobserved confounders that are stable over time. Research to date strongly suggests that patient-therapist dyads who establish a good-quality working alliance early in treatment are more likely to achieve good outcome. Moreover, the quality of the alliance in a given session is predictive of change in symptom distress by the following session. Conclusions Working alliance is predictive of psychotherapy outcome. Although later studies are stronger for drawing causal conclusions, none of the studies to date can conclusively rule out alternative explanations of alliance-outcome predictions. Whether alliance is facilitative of technical interventions or therapeutic in itself is also not clear. Future research should (1) find ways of ruling out alternative explanations of relationship variable findings, (2) study the interaction between relationship variables and technique, and (3) study the therapy relationship as a mediator of therapeutic interventions, and (4) study other relationship variables than alliance.