Learning rules are increasingly being used in macroeconomic models. However one criticism that has been levelled at this assumption is that the choice of variables for inclusion in the learning rule, and the actual specification of the learning rule itself, is arbitrary. In this paper we test how important the particular learning rule specification is by incorporating a battery of learning rules into a large-scale macro model. The model's dynamics are then compared to those from a version of the model simulated under rational expectations (RE). The results indicate that although there are large differences between the RE solution and each of the solutions under learning, differences amongst the learning rule solutions are minor