We use laboratory experiments to investigate the effect that assuming rational expectations has on structural inference in a dynamic discrete decision problem. Our design induces preferences up to the subjective rate of time preference, leaving unrestricted both this parameter and subjects' decision rules. We estimate subjects' discount rates under the assumption that all subjects use the rational expectations decision rule, and under weaker behavioral assumptions that allow decision rule heterogeneity. We find that certain sophisticated heuristics fit subjects' decisions statistically significantly better than rational expectations. However, the rational expectations assumption does not distort inferences about the cross-sectional discount rate distribution.