Mental health treatment outcome might be improved through better measurement of symptom changes and by the application of clinical prediction models. In Chapter 1, we evaluated the extent to which a depressive severity measure exhibits the implicit assumption that reductions from pre- to posttreatment that are equal to each other are of equal value. With ratings of the importance of reductions on nine symptoms from former or currently depressed patients, we found that: (a) changes on these symptoms were valued differently; (b) changes reflecting higher numeric reductions were not always judged as more important than those reflecting lower reductions; and (c) changes reflecting the same numeric reduction were valued differently depending on the level of severity at posttreatment. In Chapters 2 and 3, we assessed the potential utility of clinical prediction models within mental health. In Chapter 2, we compared the performance of a model developed using a complex algorithm with models developed using prior methodology. Using data from patients who received lithium or quetiapine for bipolar disorder, we found that a model developed using a complex algorithm predicted outcomes of patients from the held-out test set better than the models developed using prior methodology. In Chapter 3, we examined the generalizability of a clinical prediction model to mental health services within the United Kingdom. We took a “successful” clinical prediction model developed in one mental health service and used it to predict outcomes at five other mental health services. We concluded that clinical prediction models for mental health treatments can generalize to other services. We also compared the accuracy of models developed using local data with that from a model imported from another service. We found that a model built within the service it is applied to may outperform a model imported from another service. Therefore, we suggest adjusting the model for the base rate of the local population.