Background: Serious mental disorders have considerable individual and societal impact, and traditional treatments may show limited effects. Music therapy may be beneficial in psychosis and depression, including treatment-resistant cases. The challenge in systematically reviewing this field of research is that variations occur on many levels. Objectives: The aim of this review was to examine the benefits of music therapy for people with serious mental disorders. Methods: We conducted a systematic review of all prospective studies to date comparing clinically relevant outcomes of music therapy versus other conditions. Studies were combined using mixed-effects meta-analysis models, allowing to examine the influence of study design (coded as a three-level factor), type of disorder (percentage of psychotic vs. non-psychotic), and number of sessions. An iterative scheme was applied to find the appropriate weights in each statistical model; the best models were selected on the basis of their explained variance. Results: Fifteen studies (8 RCTs, 3 CCTs, 4 prepost studies) were included in the meta-analysis. All compared music therapy to standard care. Study design and type of disorder were not significantly related to outcome. Significant dose-effect relationships were identified for general, negative, and depressive symptoms, as well as functioning, with explained variance above 70%. Small effect sizes for these outcomes are achieved after 3 to 10, medium effects after 10 to 24, and large effects after 16 to 51 sessions. Dichotomous outcomes showed that the number to treat for global state is low (NNT = 1.59), indicating that one out of two patients offered music therapy will benefit from it. Conclusions: Music therapy is an effective treatment which helps people with psychotic and non-psychotic severe mental disorders to improve global state, symptoms, and functioning. Slight improvements can be seen with a few therapy sessions, but longer courses or more frequent sessions are needed to achieve more substantial benefits. This systematic review shows how mixed-effects meta-analysis models can be applied to examine the impact of several predictors, categorical as well as continuous ones, simultaneously.