The idea that teaching practices could be improved through better data systems has captured the imagination of policy makers, researchers and foundations. However, the concept of data use itself is so ill-defined as to suggest that it is more of a panacea than a coherent reform proposal. This paper examines the concept with an eye toward what we would need to know about teacher data use to ensure that it would actually yield productive improvements in teaching and would not turn out to be an ill-thought-through panacea. The paper offers a research agenda organized around two central issues: The difference between psychometric feedback and clinical feedback, and the difference between productive teaching and defensive teaching. (Contains 1 table.)