The purpose of the present investigation was to develop a construct validity approach for testing whether the separation of positive and negative item subscales is substantively meaningful in self-concept research. Results from three published studies using the Self Description Questionnaire (SDQ) III were reanalyzed. The SDQ III measures 13 distinct areas of self-concept and half of the items are negatively worded. In the reanalyses, no support was found for differentially weighting responses to positive and negative item subscales for any of the scales. Differential weighting produced little or no improvement in the prediction of self-concepts inferred by significant others (Study 1), in the short-term stability of responses (Study 2), in the consistency of responses across split-halves of each set of items (Study 3), or in the prediction of mathematical and verbal achievement (Study 3). The lack of support for the separation of positive and negative item subscales for reliability-like criteria such as stability and internal consistency suggests that support is unlikely for any validity criteria. Because scales constructed with negatively and positively worded items are a special case of the more general bipolar rating scale, the methodological approach has broad applicability for personality research. (Author/JAZ)