The authors suggest ways to improve the classification of personality disorders by changing the way the classification is developed, evaluated, and modified. Specific proposals are intended to establish a system that can be evaluated and revised so that it successively approximates a valid system. The importance of stating the principles underlying the classification is emphasized. These principles include: the conception of personality disorder underlying each diagnosis, the structure used to define diagnoses, the nature of diagnostic items, and the model for organizing diagnostic decisions. A distinction is drawn among the theoretical, measurement, and diagnostic models underlying each diagnosis. The substantive aspects of the classification are discussed in terms of the evidence required to validate diagnostic concepts. It is argued that the first step in developing a classification is to ensure the content validity of diagnostic concepts because this is a prerequisite for other components of validity. Evaluation and revision of the classification are discussed in terms of the importance of convergent and discriminant evidence. It is recommended that the classification be evaluated and revised using criteria derived from the theoretical and measurement models associated with each diagnosis. It is also recommended that the classification be evaluated in terms of the degree to which diagnostic constructs are consistent across clinicians, different sets of diagnostic exemplars, and different samples of patients. Realization of these aims should provide a classification that may be verified, modified, or disproved in the scientific tradition.