Background. Cost evaluation research in the mental health field is being increasingly recognized as a way to achieve a more effective deployment of scarce resources. However, there is a paucity of studies that seek to identify predictors of psychiatric service utilization and costs. This paper aims to critically review the published research in the field of psychiatric service utilization and costs, and discusses current methodological developments in this field. Method. Sixteen studies were identified and are critically reviewed. Results. No single variable alone can explain variations in costs between patients; instead, a range of different clinical and non-clinical variables provides a greater explanation of cost variations. Having a history of previous psychiatric service use is the most consistent predictor of higher psychiatric costs. Only one study considers indirect costs incurred by users, their families and friends and society as a whole, with the remaining 15 studies focusing on direct mental health care costs. There is a lack of studies that consider the future psychiatric service utilization and costs of care of children and older people. The cross-validation of predictive models is not yet routine, with only four of the studies including a cross-validation procedure. Conclusions. The predictive approach in mental health cost evaluation has relevance for both mental health policy and practice. However, there is a paucity of studies that focus on children, older people and indirect costs. Furthermore, there remain a number of methodological challenges to address.