The purpose of this study was threefold. First, I sought to develop and validate items with which to assess A. Bandura's (1997) theorized sources of self-efficacy among middle school mathematics students. Second, I aimed to examine the relationship between these sources, self-efficacy, relevant motivation constructs, and mathematics achievement and to investigate whether these relationships differ for boys, girls, White and African American students, and students above, on, and below grade level in mathematics. Third, through interviews conducted with students with high and low self-efficacy, and with their parents and math teachers, I sought to, examine the heuristics that students use when selecting and interpreting efficacy-relevant information. Items closely matched Bandura's description of the sources. Results from a large-scale pilot study (N = 1111) were used to develop and refine items for use in Phase 1. In Phase 1 (N = 824), a 39-item, four-factor exploratory model best fit the data. Items were again revised to strengthen psychometric properties. In Phase 2 ( N = 803), a 24-item, four-factor confirmatory factor model demonstrated the best fit. This final model was invariant across gender, ethnicity, and ability level. The items were themselves interrelated, and subscales correlated with self-efficacy, self-concept, invitations, mastery goals, and optimism across all groups. The sources were also related to teacher ratings of students' mathematics competence and to grades. No mean differences in the sources were detected by gender or by ethnicity when ability level was controlled. As hypothesized, mastery experience was the strongest predictor of self-efficacy. Results of the factor and reliability analyses revealed that the Sources of Mathematics Self-Efficacy Scale created for this study is psychometrically sound and can be confidently used to assess mathematics self-efficacy with students from Grade 6 through 8. Moreover, the nature of the items suggests that they could be easily adapted to other academic domains and used with students at all academic levels. Interview data demonstrated that students rely on combined and multiple sources when forming their efficacy beliefs. These sources included course placement, teaching structures, self-regulated learning, and cognitive self-modeling. Findings refine the tenets of A. Bandura's (1986) social cognitive theory.