This paper addresses philosophical issues concerning whether mental disorders are natural kinds and how the DSM should classify mental disorders. I argue that some mental disorders (e.g. schizophrenia, depression) are natural kinds in the sense that they are natural classes constituted by a set of stable biological mechanisms. I subsequently argue that a theoretical and causal approach to classification would provide a method for classifying natural kinds that is superior to the purely descriptive approach adopted by the DSM since DSM-III. My argument suggests that the DSM should classify natural kinds in order to provide predictively useful (i.e. projectable) diagnostic categories and that a causal approach to classification would provide a more promising method for formulating valid diagnostic categories.