Abstract In a 2 × 2 factorial design, goal valence, and assumed similarity to someone better off than oneself were manipulated and locus of social comparison measured. Comparison with someone better off than oneself was found to be more frequent under (a) conditions of positive than of negative goal valence, and (b) conditions of low than of high assumed similarity. These results appear partially explicable in terms of two postulated ends served by the social comparison of abilitics—one self-evaluative and the other self-enhancing. Data on degree of interest expressed in the scores of others provide independent support for this interpretation. The major difference between Festinger's and the present conceptualization involves the role of similarity in mediating self-evaluation. Within reasonably broad limits, it appears to be the positive instance, rather than the similar one, which affords accurate self-evaluation.