Increased attention to the issue of aid effectiveness has sparked a flurry of empirical studies attempting to measure the macro-level impact of aid flows on the performances of developing countries. These studies yield ambiguous and even contradictory results and one possible source of confusion is the fact that they are not grounded in solid theorizing. This paper tries to remedy this lacuna by proposing a principal-agent model in which the donor monitors the use of aid and metes out sanctions in the event of fraud detection. Its most original feature lies in the assumed comparability between internal (domestic) and external (donor-imposed) disciplines and the resulting possibility of studying the behaviour of aggregate discipline. We show that, contrary to intuition, an (exogenous) improvement of domestic discipline may be over-compensated by the donor so that total discipline actually decreases and elite capture paradoxically increases. This implies that the relationship between domestic and total disciplines may be non-monotonous. We highlight the crucial role of the shape of the cost functions to obtain not only the above paradox but also corner solutions in which the donor optimally chooses to refrain from imposing any external discipline. The central lesson to draw from the whole exercise is therefore that no simple general testable prediction can be inferred from economic theory regarding the impact of aid on the donors’ objective even when the quality of domestic governance or the policy environment is considered.