A theory is expanded which attempts to explain why some predictors are more valid than others. In essence, high validity is expected when a measure accurately reflects the degree to which work performance depends upon three domains of characteristics: (1) universals, the characteristics required by all work; (2) occupationals, the characteristics required by certain jobs; (3) relationals, the characteristics needed to relate to others in specific settings. It is suggested that selectors have frequently ignored the third domain.The paper considers the theory from the viewpoint of criteria and concludes by outlining the uses of the theory in terms of (1) further theory building; (2) improving understanding of existing predictors; (3) improvement of existing predictors; and (4) development of new predictors.