We define and model research production at Embrapa, the major Brazilian institution responsible for applied agricultural research. The main theoretical framework used is Data Envelopment Analysis — DEA. The economic interpretation of these models is explored to assess scale, congestion and cost efficiencies. Efficiency results are used to test for differences among types of research units and for the scale of operation. A further analysis of agricultural research in Brazil is carried out with the inclusion of three research centers in Argentina. Finally, DEA estimates are compared with the fit of a stochastic frontier.