The purpose of this study was to further clarify the structure of social persuasion as a source of self-efficacy in early adolescence and to examine the influence of social persuasion on STEM self-efficacy. Specifically, we proposed that social persuasion for math should be considered a multifactor construct for middle school students when predicting self-efficacy for STEM courses. Data were collected from 1,445 middle school students using a modified measure of social persuasions developed by Usher and Pajares (2009) and self-efficacy for STEM courses developed by Hackett and Betz (1989). Using factor analysis followed by structural equation modeling on two randomized samples, our findings indicate that family, peer, and courses/career persuasion in math are significant predictors of STEM courses self-efficacy, but not teacher persuasion. Implications and directions for future research are discussed.