In this paper the authors extend recent work on mediation in a multilevel setting and on causal inference under interference among units to develop a template for the mediation analysis of group randomized educational interventions. The present work will contribute to the literature on interference, in particular on interference in the context of mediation analysis. They will show that not only does the total effect of the intervention decompose into a direct effect and an indirect effect mediated through classroom quality but also that the indirect effect itself decomposes into an effect mediated through the quality of a child's own classroom and a spillover effect from the quality of the other classrooms at a school. They will give some results for the identification of these direct, indirect and spillover effects and consider the consequences of ignoring interference when it is in fact present. They will then analyze the effects of the Reading, Writing, Respect and Resolution (4Rs) intervention in a group randomized trial. (Contains 1 table.)