This article investigate the sensitivity of a simple experiment in Agent-based Computational Economics(ACE) to the choice of representation of the agents. We demonstrate that significant differences in experimental outcomes result from varying the representation, even when other experimental factors are left as close to constant as possible. All experimental parameters other than agent representation and, necessarily, variation operators are left constant. In learning the classic Divide the Dollar game proposed by Nash the agents exhibit distinct behaviors. The implications for requiring that representations be normed against experimental results obtained with human agents are discussed.