Modern classifications of mental disorders containing operationalized diagnostic criteria have had an important beneficial effect on interrater reliability. From the beginnings of modern approaches to psychiatric diagnosis (8—10), however, many investigators have attempted, largely without success, to establish validity for the existing disorders as "natural kinds" or to refine or subdivide them in a search for homogeneity. Over time, instead of validation, severe problems with the existing classifications have become increasingly obvious. These include not only the co-morbidity problem already discussed, but also the need for clinicians to rely with uncomfortable frequency on "not otherwise specified," or NOS, diagnoses (11, 12). The need to rely on NOS diagnoses suggests that the DSM system has a fundamental problem of overspecification: patients fully meeting criteria for DSM-IV disorders, especially for a single DSM-IV disorder, may well be the minority of individuals seen in clinical settings. To compound these problems so evident to clinicians, scientific studies, most notably genetic studies, have found shared genetic risk factors across putatively distinct disorders. These shared risk factors have been identified both by twin studies that identify aggregate genetic risk (3) and, increasingly, by molecular studies, which, for example, have found shared genetic contributions to schizophrenia and bipolar disorder (13).