Based on study of (a) the HIPC debt relief initiative, (b) the linking of aid to policy conditionality and (c) transactions-cost arguments in favour of programme aid, this article argues that major elements in the new aid agenda may not be well-based empirically. This is partly because of inadequate knowledge, but particularly because the evidence often conflicts with political preferences. As a result, it is likely that large amounts of aid resources are being misdirected. Ways are suggested of narrowing the gap between evidence-based and ‘political’ decisionmaking. In the meantime, donors should avoid diverting more aid into debtrelief, should roll back their reliance on policy conditionality, and should exercise pragmatic caution in the expansion of programme assistance.