In this paper we use novel data to test the direct and indirect paths between teacher self-efficacy and student outcomes. This includes how teacher self-efficacy is linked to student, teacher, and expert rater views of lesson quality. Our results illustrate how the link between teacher self-efficacy and instructional quality is sensitive to how lesson quality is measured, with large effects when based on teacher reported outcomes but no association when based on the ratings of expert observers. Virtually no relationship is found between teacher self-efficacy and student outcomes. We thus conclude that while there is probably some positive association between teacher self-efficacy and the quality of their instruction, the strength of this relationship is relatively weak.