For more than a decade now, so-called «evidence-based policies» (Ebp) have become a leading trend in public action. Drawing on three complex epistemological notions developed by Amartya Sen (i.e. «informational basis of judgment in justice», «description as choice», and «positional objectivity »), the paper argues that Ebp are based on specific and questionable cognitive and normative bases that to a large extent shape what is afterwards labeled as «evidence». Instead of taking for granted their claim to objectivity and efficiency, Ebp ought to be submitted to three sets of questions, namely: what kind of information is labeled as evidence? who makes decisions about this? what information is discarded or left aside in the process of defining evidence? The conclusion insists on the political implications of the reflection developed in the paper, esp. the necessity to move from Ebp to more genuinely democratic policy-making. only subscribers can see the full article