Imagine you are playing chess. As you think about your next move, you consider the outcome each possibility will have on the board, and the likely responses of your opponent. Your knowledge of the board and the rules constitutes an internal model of the chess game. Guiding your behavior on the basis of model-predicted outcomes of your actions is the very definition of cognitive planning. It has been known for many decades that humans and animals can plan (Tolman, 1948), but the neural mechanisms of planning remain largely unknown. Recently, a powerful new tool for the study of planning has become available: the ‘two-step’ task introduced by Daw et al. (2011). This task allows, for the first time, the collection of multiple trials of planned behavior within a single experimental session, opening the door to many new experimental possibilities. We have adapted the two-step task for use with rodents, and developed a semiautomated pipeline to efficiently train large numbers of animals. Here, we show that the rodent two-step task reliably elicits planning behavior in rats, and we characterize the role of the orbitofrontal cortex (OFC) in this planning behavior. We find that inactivations of OFC substantially impair the ability to plan, and that single units in OFC encode planning-related variables, such as the values associated with actions taken at each step in the two-step task. These data demonstrate the OFC is crucial for planning, and begin to shed light on the computational role that it plays in the planning process.