The Government of Zimbabwe's (GOZ) emergency social safety nets have been credited with preventing famine and protecting livelihoods during three consecutive drought years in the 1980s, most notably by Dreze and Sen. Zimbabwean authors have been more sceptical about GOZ's social safety nets (SSNs) and have pointed out both that Zimbabwe's drought-triggered SSNs have not solved food insecurity problems and that GOZ's other, regular SSNs have been far less effective. Using definitions of poverty based on income, consumption, assets and capabilities, this thesis asks whether GOZ's social safety nets - regular and emergency - have provided poor people with relevant, timely and adequate assistance in the 1990s. I explore this question using two puzzles in modem political economy. The first is the assertion by Comia and Stewart that SSNs face a trade-off between errors of failing to reach the target group (F-mistakes) and errors of excessive coverage of non-target groups (E-mistakes). The second is complex redundancy. While the new public management seeks more efficient, leaner administrative forms for public services, complex redundancy suggests that a certain amount of duplication promotes greater programme effectiveness, and a greater ability to absorb shocks. The thesis uses eight nationwide household surveys conducted by GOZ 1992-1997, as well as administrative data and qualitative interviews with SSN administrators, applicants and beneficiaries. The methodological contribution of this thesis is to show the usefulness (and limitations) of secondary analysis of existing household surveys to study SSNs in Zimbabwe, a topic previously explored almost entirely through administrative data and qualitative interviews. The thesis questions whether there is a simple trade-off between E- and F-mistakes in SSNs. The trade-off exists in mature or stable SSNs, but life-cycle, political and managerial factors also affect error patterns. The life-cycle is particularly important for emergency SSNs. Economists have overemphasised the formal design features of SSNs as the determinant of the error pattern. While confirming that Dreze and Sen's favourable overall assessment of Zimbabwe's emergency SSNs continued to be valid in the 1990s, this thesis suggests a need to modify their view of Zimbabwe as an exemplar of effective public action against hunger. In addition to emphasising anti-poor shifts in sectoral policies in the 1990s, the thesis stresses the need to disaggregate analysis of SSNs by programme and time period, since there have been large differences in levels of effectiveness between the various emergency SSNs, between the emergency and regular SSNs, and over time within a given SSN. Many predictions of complex redundancy theory are borne out, but some room remains for improving efficiency without reducing effectiveness. Zimbabwe's SSNs generally have a slight pro-poor bias in that the poor are more likely to get benefits than the non-poor, whether poverty is defined in terms of income, consumption, assets or capabilities. But F-mistakes are still high for Child Supplementary Feeding, Public Assistance, the tillage component of the Agricultural Recovery Programme, and the Social Development Fund. E-mistakes are also significant in some cases, notably the Social Development Fund. The thesis confirms Dreze and Sen's assertion that the state can play a large and positive role in ensuring social security, even in poor countries with stagnant economies. However, emergency SSNs seem to be easier to manage effectively than regular SSNs, especially if they are administratively simple, universalistic and of short duration. The current political turmoil in Zimbabwe does not, however, bode well for effective implementation of SSNs in the next decade.