I estimate the effect of peers, defined as teachers at the same school and grade level, on the own effectiveness of teachers, measured by value-added scores. Traditional estimates using leave-out means imply significant and large positive spillovers among teachers. In this paper, I exploit a more compelling research design following Mas and Moretti (2009), which approximates the following thought experiment in a regression framework: A low value-added teacher, Teacher B is randomly replaced by a new, high value-added teacher, Teacher C, at a particular school-grade level. How does the value-added of incumbent Teacher A, who worked at the same school-grade level the year before, change in response? Among North Carolina elementary school teachers, I find that in response to the replacement of a teacher by a 1 standard deviation (SD) better teacher, incumbent peers' value-added increases by about 0.12 SDs, which is about one half of traditional estimates. Moreover, I uncover large heterogeneities across schools, which shed light on how much of these estimates are due to pure peer effects among teachers versus student sorting. Using the classification of schools developed in Horváth (2015), I find that the results are driven by schools that sort students to teachers, while estimates are low and insignificant in random assignment schools. Looking at changes in observed student characteristics in the incumbent teachers' classrooms reveals that assignments in random schools do not change when a colleague is replaced, just as expected. Therefore, the insignificantly small estimates in these schools provide clean estimates of peer effects. In contrast, average prior achievement in incumbent teachers' classrooms significantly decreases in sorting schools. In these schools, therefore, student sorting may explain the spillovers among teachers.