The thesis looks at the macroeconomic impact of foreign aid. It is specially concerned with aid's impact on the public sector of less developed countries < LDCs> . Since the overwhelming majority of aid is directed to the public sector of LDCs, one can only understand the broader macroeconomic impact of aid if one first understands its impact on this sector. To this end, the thesis econometrically estimates " fiscal response" models of aid. These models, in essence, attempt to shed light on public sector fiscal behaviour in the presence of aid inflows, being specially concerned with the way aid is used to finance various categories of expenditures. The underlaying concern is to extent to which aid is " fungible" -that is, whether it finances consumption expenditure and reductions in taxation revenue in LDCs. A number of alternative models are derived from a utility maximisation framework. These alternatives reflect different assumptions regarding the behaviour of LDC public sectors and relate to the endogeniety <as opposed to exogeniety> of aid, whether or not recurrent expenditure is financed from domestic borrowing and the determination of domestic borrowing. The original frameworks of earlier studies are extended in a number of ways, including the use of a public sector utility function which is fully consistent with expected maximising behaviour. Estimates of these models' parameters are obtained using both time-series and cross-section data, dating from the 1960s, for Bangladesh, India, Pakistan and the Philippines. Both structural and reduced-form equations are estimated. Results suggest that foreign aid <defined as all foreign inflows to the official sector> is indeed fungible, albeit at different levels. Moreover, the overall impact of aid <both loans and grants> on public sector investment, consumption, domestic borrowing and taxation varies between countries. Generally speaking, aid leads to increases in investment and consumption expenditure, but reduces taxation and domestic borrowing. Comparative analysis does, however, show that these results are highly sensitive to alternative behavioural assumptions and, therefore, model specification.