Pharmacogenetic risk models offer great promise as treatment decision tools; however, their uptake in routine clinical practice is so far disappointing, not least due to the lack of evidence of their benefit in randomized controlled trials and other types of studies. Prior to conducting such a study, it is imperative that the model's predictive capability is first of all proven, and that it is shown to be superior to the most appropriate alternative model. When demonstrating predictive capability, clinical implications of applying the model should be a key consideration, and the Decision Curve Analysis method takes this into account for binary outcomes. Furthermore, when comparing a novel model to the best alternative, methods such as Net Reclassification Improvement or Integrated Discrimination Difference are recommended as they provide a more reliable comparison than other methods currently in common use. Where outcome is continuous, such as therapeutic dose, assessing a model's performance is generally more intuitive and straightforward since the aim is to achieve a predicted dose as close as possible to the true therapeutic dose.