There is growing concern among researchers and governmental officials that knowing what works in education is important, but not enough for school improvement. Sound evidence alone is not sufficient for large-scale, sustainable change, both because practitioners may consider it irrelevant to their own problems of practice or run into challenges when they try to implement. Failed attempts at replicating positive outcomes in new (or simply expanded) settings underscore the need for a different relationship between research and practice, one that takes a systemic perspective on improvement and transforms the role for research. In this article, we describe the new science of improvement and where it sits in the evolution of research on education policy implementation. We discuss the roots of the approach as well as its key features. We explain how the work differs from that of traditional research and end with illustrations of this difference from our experiences with the National Center on Scaling Up Effective Schools. [For the corresponding grantee submission, see ED595047.]