We conduct a laboratory experiment to evaluate the econometric procedure for estimating dynamic games proposed by Aguirregabiria and Magesan (2012, AM hereafter).  AM’s procedure is valuable because it relaxes the equilibrium assumption that players’ beliefs accurately represent the actual decisions of their opponents.  The advantage of our approach over conventional Monte Carlo studies is that our data are generated by human participants rather than simulated agents, without losing the availability of “known” parameters for evaluation purposes.  We (i) discover the conditions under which AM’s procedure provides reliable inferences with respect to the structural parameters, and (ii) draw inferences about the nature of human players’ beliefs in these environments.