Signalling models are studied using experiments and adaptive learning models in an entry limit pricing game. Even though high cost monopolists never play dominated strategies, the easier it is for other players to recognise that these strategies are dominated, the more likely play is to converge to the undominated separating equilibrium and the more rapidly limit pricing develops. This is inconsistent with the equilibrium refinements literature (including Cho-Kreps’ intuitive criterion) and pure (Bayesian) adaptive learning models. An augmented adaptive learning model in which some players recognise the existence of dominated strategies and their consequences predicts these outcomes.