Described is a study investigating the feasibility and predictive value of the Teacher Feedback Coding System, a novel observational measure of teachers’ feedback provided to students in third grade classrooms. This measure assessed individual feedback events across three domains: feedback type, level of specificity and affect of the teacher. Exploratory and confirmatory factor analysis revealed five factors indicating separate types of feedback: positive and negative academic-informative feedback, positive and negative behavioral-informative feedback, and an overall factor representing supportive feedback. Multilevel models revealed direct relations between teachers’ negative academic-informative feedback and students’ spring math achievement, as well as between teachers’ negative behavioral-informative feedback and students’ behavior patterns. Additionally, a fall math-by-feedback interaction was detected in the case of teachers’ positive academic-informative feedback; students who began the year struggling in math benefitted from more of this type of feedback. Finally, teachers’ feedback was investigated as a potential mediator in a previously established relation between teachers’ self-reported depressive symptoms and the observed quality of the classroom environment. Partial mediation was detected in the case of teachers’ positive academic-informative feedback, such that this type of feedback was accountable for a portion of the variance observed in the relation between teachers’ depressive symptoms and the quality of the classroom environment.