A Markov regime switching model for exchange rate fluctuations, with time-varying transition probabilities, is used in constructing a monthly model for predicting currency crises in Southeast Asia. The approach is designed to avoid the estimation inconsistency that might arise from misclassification errors in the construction of crisis dummy variables which other approaches (such as probit/logit and signaling) require. Our methodology also addresses the serial correlations and sudden behavior inherent in crisis occurrence, identifies a set of reliable and observable indicators of impending crisis difficulties, delivers forecast probabilities of future crises over multi-period forecasting horizons, and offers an empirical framework for analyzing contagion effects of a crisis. Our empirical results indicate that the Markov switching model is moderately successful at predicting crisis episodes, but also points to future research in various directions. Most early warning systems for currency crises have used either probit or signaling. Several issues can be raised regarding these techniques: the need for a priori dating of crisis occurrence, the use of arbitrary thresholds, inadequate modeling of the dynamics in the system, among others. We present an alternative framework, based on a Markov-switching model of exchange rate fluctuations with time-varying transition probabilities, which addresses these concerns.