Declining land availability and increasing rural poverty hamper further production growth and agricultural development in Burundi. Yet, over the last decades, food demand continued to increase as a result of high population growth and urbanisation. In a country like Burundi, where it is estimated that 90% of the people live from agriculture for both food supply and income earnings, agricultural specialists urge for a more optimal use of resources in agricultural production as a pathway out of poverty. However, it is difficult to optimize the complex prevailing farming systems without radical changes in farmers’ decision making. Over time, small-scale farmers have developed complex cropping systems that are both intensive and risk averse in response to acute land scarcity and unreliable food markets. This has resulted in high rates of self-reliance among farmers and highly diversified mixed cropping systems. This paper examines the potential and effects of agricultural specialization in optimizing land use and farm production. We use a farm level mathematical programming model which is applied to different farm types. A special feature in this paper is the consideration of minimal household food needs. Simulation results revealed that optimal farm plans differ by farm type as each of them is subject to specific asset endowment constraints. Farmers could improve land use by producing a specific range of crops and benefiting from economies of scale; large farms can significantly increase returns to land, while landless households can benefit from increased returns to labour. However, the reforms require better markets, storage facilities and credit supply mechanisms.