This paper argues that the use of "laboratory" data sets can add substantially to the teaching of production economics at the graduate level. Optimal experimental designs for generating pseudo data from a process model are discussed. These are shown to depend of the functional form to be estimated. We choose the translog form for our multiproduct profit function and compare alternative approaches to estimation, using pseudo data from a farm-level linear programming model. Particular restrictions on this profit function are also considered. Finally, aggregation of output prices is shown to alter substantially input price elasticities of demand.