Therapeutic alliance is an umbrella term referring to core aspects of the interaction and relationship between patient and practitioner during treatment. It has long been considered an important component of success in psychological and medical treatments. A survey of practitioners in child mental health (Kazdin, 1997) found that 95% thought that the relationship with the patient was the most important predictor of treatment outcome; there is research evidence suggesting the significant impact of alliance quality on outcome in adults and children, for both psychological (Martin et al., 2000, Shirk and Karver, 2003; Shirk, Karver and Brown, 2011) and general medical (Burkitt-Wright et al., 2004) treatments. Alliance, however, has been relatively little researched in childhood and until recently the emphasis (in both research and training) has been much more on the protocol details of treatment methods as opposed to detailed understanding of treatment process and the practitioner-patient relationship. Studies reporting associations between therapeutic alliance and treatment outcome have often been weakened by methodological difficulties in measurement and have failed to settle the direction of causality between symptom change and alliance (Kazdin and Nock, 2003). In treatment trials, alliance is often only measured in the experimental arm; this makes analysis of its effect difficult (Dunn and Bentall, 2007, and Emsley et al., 2010).This study represents an exceptional opportunity to address these limitations. It makes use of data collected during one of the most rigorous recent studies done in child mental health in the UK (Goodyer et al., 2007). This enables detailed study of the therapeutic relationship during treatment and allows testing of the effects of this relationship on the success of treatment. Sessional audiotapes were available within both arms of this trial. Purposeful selection of tapes from both arms of the trial during treatment were transcribed and rated for treatment alliance. Other data already collected in the trial was included in an analysis to address questions of direction of causality of alliance in relation to symptom change during treatment and the way that alliance may explain treatment effect heterogeneity.The results indicate a complex effect of alliance upon outcome. There is a relationship between early alliance score and clinical improvement, but the relationship is not straightforward and the predictive effect of alliance appears to depend on differences in patient groups and therapist effects. Analysis of treatment effect heterogeneity suggests that therapeutic alliance is associated with the individual treatment effect and implies that with poor alliance, more treatment may be detrimental. The complexities of the results are discussed with reference to implications for further research in this area as well as clinical practice.