The therapeutic value of alliance is a contested supposition. Although many theorists and researchers believe that alliance is therapeutic in itself, others see it as a byproduct of effective treatment or as a common nonspecific factor enabling the truly effective ingredients of treatment to work. For many years, the debate was confined mainly to the domain of theory, and no studies were available to confirm which of these approaches is correct. The only empirical evidence that existed was studies showing a correlation between alliance and outcome, and advocates of the above conflicting opinions used the same correlation to prove the validity of their position. Over the last few years, however, a revolution has taken place in alliance research, which brings this theoretical debate into the realm of the empirical. Several recent alliance studies have applied advanced methodologies to achieve this aim. Based on an integration of these studies, the present article proposes a new model for understanding the potential therapeutic role of alliance as sufficient to induce change by itself. The model stresses the importance of differentiating between patients' general tendencies to form satisfying relationships with others, which affect also the relationship with the therapist ("trait-like" component of alliance), and the process of the development of changes in such tendencies through interaction with the therapist ("state-like" component of alliance). The former enables treatment to be effective; the latter makes alliance therapeutic. Based on the literature, this article attempts to determine which of these components is the predictor of treatment outcome.