`titrebPredicting Banking Crises: A Robust Early Warning System`/titreb An Early Warning System aims at predicting banking crises in 48 countries from 1977 to 2010. Bayesian Model Averaging is used to get robust predictions. The growth rates of economic activity and credit, financial liberalization, and the external indebtedness are decisive signals of banking crises. Minimizing a predictive loss function, we find an optimal rate of false signals and missed crises that improve the insample forecasts before 2006. As a sign of its robustness, this warning system helps signal 49 of the 52 registered banking crises since 2006 despite heterogeneities across countries.