The authors advance an argument that placing observation of actual teaching as a central feature of accountability frameworks, teacher preparation, and basic science could result in substantial improvements in instruction and related social processes and a science of the production of teaching and teachers. Teachers' behavioral interactions with students can be (a) assessed observationally using standardized protocols, (b) analyzed systematically with regard to sources of error, (c) validated for predicting student learning, and (d) changed (improved) as a function of specific and aligned supports provided to teachers; exposure to such supports is predictive of greater student learning gains. These methods have considerable promise; along with measurement challenges, some of which pertain to psychometrics, efficiency, and costs, they merit attention, rigorous study, and substantial research investments.