The paper discusses the impact of centre selection on the generalisability of randomised controlled trial (RCT)-based economic evaluations and suggests a future research agenda. The first section briefly reviews the current methods for addressing generalisability. We argue that these methods make no verifiable assumptions about how representative the recruiting centres are to the population of centres in the jurisdiction. The second section uses data from a multicentre RCT to illustrate that cost-effectiveness estimates can be influenced by the sample of recruiting centres. Finally, we propose two concepts that may advance generalisability research. First, we distinguish between the 'research space' and the 'policy space' and argue that policy makers are interested in the latter, while current methods describe the former. Second, we propose a centre-specific generalisability index used at RCT design stage to address generalisability. We conclude that future research should focus on generalisability at RCT design stage rather than on post hoc analyses.