Several recent papers report evidence of an apparent statistical bias in inflation expectations and interpret these findings as overturning the rational expectations hypothesis. In this paper, we investigate the validity of such an interpretation. We present a computational dynamic general equilibrium model capable of generating aggregate behavior similar to the data along several dimensions. By construction, model agents form "rational" expectations. We run a standard regression on equilibrium realizations of inflation and inflation expectations over sample periods corresponding to those tests performed on actual data and find evidence of an apparent bias in inflation expectations. Our experiments suggest that this incorrect inference is largely the product of a small sample problem, exacerbated by short-run learning dynamics in response to infrequent shifts in monetary policy regimes.