The Bertrand paradox describes a situation in which two competing firms reach an outcome where both price at marginal cost. In laboratory experiments, this equilibrium is not generally observed. Existing empirical works on Bertrand competition have found evidence for boundedly rational models. We find that such models are useful in organizing behavior in early stages of the game, but less so in later stages. We show that a new model, coarse grid Nash equilibrium, based on the assumption that subjects discretize the strategy space, explains the data better.