The 'right' choice of instruments and modalities to provide aid to developing countries in support of poverty reduction and economic development is arguably the most contested issue in the current international debate on aid effectiveness. A particular controversy exists around the provision of aid in the form of budget support to avoid high transaction costs and other shortcomings of traditional project-based aid. Critics argue that this kind of 'programme aid' involves unacceptably high fiduciary risks due to the fungibility of budgetary funds. A more recently proposed form of aid is in the form of results-based aid or aid on delivery. Proponents argue that this provides donors with better control over the use of aid resources. This paper demonstrates in a simple principal-agent framework with asymmetric information that in the absence of transaction costs, for a wide range of combinations of aid dependency and recipient government commitment to reduce poverty, all three forms of aid are equivalent with regard to fungibility and fiduciary risks. The paper proceeds to demonstrate that as long as donors can rely on the recipient government to be at least minimally committed to poverty reduction, a well co-ordinated modality mix of general budget support and aid on delivery does not bear higher fiduciary risks than project aid. It concludes that if project aid does indeed involve higher transaction costs than budget support, donors should provide aid in the form of such a modality mix, albeit only if they are able (and willing) to closely co-ordinate their support.