The construct validity of four self-concept (SC) traits (general SC, academic SC, English SC, mathematics SC), as measured by three different scales (Likert, semantic differential, Guttman) for low- (n=252) and high-track (n=588) Canadian high school students, was assessed using both the Campbell-Fiske criteria, and a comparison of hierarchically nested covariance structure models. Confirmatory factor analysis was used to model hypotheses related to convergent and discriminant validity and to test directly equivalencies of traits and methods. Findings indicate that assumptions of invariant construct validity cannot be taken for granted; differences in both the measurement and structure of SC were found. Academic SC, as measured by the Likert and Guttman scales, was problematic for the high track. These scales appeared to elicit different types of responses from high and low ability students. Tests of invariance formally confirmed this result. Discriminant validity of the trait factors was also less clear for the high track, but this may have been a measurement problem. Method bias was clearly more evident for the high than for the low track. Method bias effects for each scale type, as well as all but one trait correlation, were found to be noninvariant. A 5-page list of references and eight tables are included. (LPG)