This article deals with the selection of items in a self-report test that measures a broad trait. There is an inherent conflict between selecting items that maximize internal consistency and ones that contribute to breadth. Alternative procedures, including criterion keying, combining homogeneous subscales, and construction of scales with divergent items with modest relations to the whole scale scores, were discussed. Although not included among recommended procedures by modern test theory, the latter approach has much to recommend it under appropriate circumstances. The discussion of psychometric principles was generalized to the person-situation debate.