Using student examination data linked to longitudinal teacher personnel data, we document that a teacher’s students have larger test score gains when she experiences an improvement in the observable characteristics of her colleagues. Using within-school and within-teacher variation, we further show that a teacher’s students have larger test score gains when she has more effective colleagues (based on their own students’ achievement gains from an out-of-sample pre-period). A one standard deviation increase in average teacher peer quality is associated with an increase of 0.02 and 0.04 standard deviations in student test score growth in reading and math respectively (about one third of the own-teacher effect). These spillovers are strongest for less experienced teachers, persist over time and teachers perform best when they are the weakest of their peer group– suggesting a peer learning interpretation. Consistent with this interpretation, conditioning on historical peer quality reduces the explanatory power of individual teacher effects by twenty percent.