A common design in education research for interventions operating at a group or cluster level is a cluster randomized trial (CRT) (Bloom, 2005). In CRTs, intact clusters (e.g., schools) are assigned to treatment conditions rather than individuals (e.g., students) and are frequently an effective way to study interventions because they permit researchers to accommodate existing school structures and the multilevel nature of many educational interventions. In this study, the author developed closed form analytic expressions for the test statistics of the mediation effect in three level CRT designs examining cross-level (e.g., 3-2-1) mediation. Using this test statistic, he estimated the power to detect cross-level mediation effects under different scenarios to assist researchers in the planning and design of multilevel mediation studies. To fix ideas, consider an example where the intervention is a professional development program to increase pedagogical knowledge and it is assigned at random to intact schools. In theory, exposure to the professional development program (treatment) should increase teachers' pedagogical content knowledge (mediator) which in turns increases students' achievement levels (outcome). The study could be described as a 3-2-1 cross-level mediation design because its hypothesized that a treatment assigned at the school level (level 3) impacts a classroom level (level 2) mediator which in turn impacts a student level (level 1) outcome. The focus of this study was to develop a framework to assess the power to detect cross-level mediation for this particular three level CRT design. A figure is appended.