Although the construct of self-monitoring has assumed a central role in the description and explanation of human social behavior, there is considerable disagreement about the performance of the Self-Monitoring Scale, the primary instrument for measuring individual differences in self-monitoring. The scale was originally designed to tap into a single source of variance; however, factor analyses have consistently demonstrated that the scale is factorially complex. Our review of published factor analyses of the Self-Monitoring Scale revealed evidence consistent with several different latent structures underlying responses to the scale's 25 items. In order to clarify the latent structure of the scale, we statistically compared, using the covariance structure model, first- and second-order measurement models suggested by previous factor analyses. We conducted parallel analyses for the original 25-item scale and a recently introduced 18-item version of the scale. None of the hypothesized models provided a compelling fit to the sample data (N= 1,113). The most favorable results were in support of a first-order measurement model with 18 indicators of three intercorrelated first-order latent variables. Moving from confirmatory to exploratory model testing, we were able to improve the fit of the best-fitting model by specifying a fourth latent variable indicated by items with which high self-monitors were expected to disagree. We discuss the implications of these results for the conceptualization of self-monitoring and for the role of factor analysis in the accumulation of knowledge about theoretical constructs.