Regime-switching rational expectations models, in which the parameters of the model evolve according to a finite state Markov process, have properties that differentiate them from linear models. Issues that are well understood in linear contexts, such as equilibrium determinacy and stability under adaptive learning, re-emerge in this new context. This paper outlines these issues and defines two classes of equilibria that emerge from regime-switching models. The distinguishing feature between the two classes is whether the conditional density of the endogenous state variables depends on past regimes. An assumption on whether agents condition their expectations on past regimes has important implications for determinacy and equilibrium dynamics. The paper addresses the stability properties of the different classes of equilibria under adaptive learning, extending the learning literature to a non-linear framework.