The role of agriculture in development remains much debated. This paper takes an empirical perspective and focuses on poverty, as opposed to growth alone. The contribution of a sector to poverty reduction is shown to depend on its own growth performance, its indirect impact on growth in other sectors, the extent to which poor people participate in the sector, and the size of the sector in the overall economy. Bringing together these different effects using cross-country econometric evidence indicates that agriculture is significantly more effective in reducing poverty among the poorest of the poor (as reflected in the $1-day squared poverty gap). It is also up to 3.2 times better at reducing $1-day headcount poverty in low-income and resource rich countries (including those in Sub-Saharan Africa), at least when societies are not fundamentally unequal. However, when it comes to the better off poor (reflected in the $2-day measure), non-agriculture has the edge. These results are driven by the much larger participation of poorer households in growth from agriculture and the lower poverty reducing effect of non-agriculture in the presence of extractive industries.