This article details the design of an impact evaluation of the Millennium Villages Project in Northern Ghana. The evaluation is particularly challenging because the intervention cannot be randomised; it is clustered in a group of homogeneous communities and likely to generate spill-over effects. We propose a difference-in-differences design selecting control communities based on a propensity score and collecting five rounds of yearly data. We address a number of evaluation questions in relation to testing the breaking of the poverty trap, assessing project externalities, the role of qualitative research, cost-effectiveness and project synergies, sustainability and scalability in the presence of scale economies. © 2013 Copyright 2013 Taylor & Francis.