The orbitofrontal cortex (OFC) has long been considered critical for value-based decision making, but its precise role is a point of contention. One hypothesis is that OFC computes subjective value and drives economic choice [1]. A second hypothesis is that OFC represents a cognitive map or state space (i.e., a representation of the different states in a task and the transitions between them) [2]. We trained rats on a novel temporal wagering task with partially hidden states (blocks of trials with low or high rewards). Rats must determine how long to wait for a reward, providing an explicit behavioral readout of subjective value. Rats’ wait times are modulated by both the offered reward volume and the hidden block. Bilateral muscimol inactivation of lateral OFC (lOFC) reduces modulation of wait time by block but does not impair modulation of wait time by reward volume. This suggests that lOFC is necessary to infer the current block based on knowledge of the task structure but is not required to compute subjective value, per se. We extend these findings using behavioral modeling to address how lOFC contributes to inference. We fit a behavioral model that uses Bayes’ Rule to predict the identity of the current block. The model includes parameters representing the opportunity cost in each block (which dictates the wait time in different blocks), and a parameter capturing the extent to which rats use an optimal prior, which contains knowledge about block length and transition probabilities. Results suggest that rats use a less informative prior when lOFC is inactivated, but other parameters are not affected. Electrophysiological recordings from lOFC reveal encoding of hidden states (blocks). These data suggest that lOFC promotes the use of a prior that incorporates knowledge of the task structure for inferring partially observable states of the environment.