This article uses the concept of "boundary object", first developed within science studies by Star and Griesemer, to analyse curriculum policy implementation. It employs as a vehicle a significant but contested reform of the science curriculum in schools in England from 2006 onwards, drawing empirically on an extended study of the reform, using public documentation and fieldwork in schools. The focus of the article is on the processes of mediation which are invoked during the implementation process. Star and Griesemer characterized boundary objects as entities which are shared across institutional and other social boundaries, but are sufficiently flexible and indeterminate to satisfy diverse agendas. In this study a curricular element called How Science Works is constituted as a boundary object. Its implementation is set within a network of institutions with different imperatives. The overall effect is to distribute the implementation process, and localize the meaning of the reform. This in turn enables what Star called "co-operative action in the absence of consensus". Complementing and sometimes working against this are mechanisms of accountability dependent on public information. These create pressures for standardization of practices, and thus of meanings, which can both undermine the working of the network and lead to reification of professional practices. The article concludes with some reflections on the implications of this analysis for curriculum developers. (Contains 17 notes and 1 figure.)