We describe a new approach to the use of video-based technology for conducting controlled experiments in classroom contexts. Specifically, we describe a process for editing video recordings of live classroom lessons to create multiple versions, such that only one aspect of the lesson is systematically varied. Other aspects of the instruction are all held constant, including audio, curricular content, student participation, and other notoriously hard-to-control details of the interactional context that impact learning (e.g., gestures, affect). These edited lesson versions can be randomly assigned to students by condition within classrooms to meet a high standard for random assignment. This technology provides opportunities for deriving causal data on the efficacy of teaching practices through stimuli approximating a typical everyday classroom context.