This study reviews the methodological basis of early warning indicators and suggests that most financial crises are potentially predictable on the basis of both theory and empirical evidence. An early warning system is simply a statistical model which forecasts crises. Thus both the problems and the experiences associated with economic forecast modeling are relevant. In practice, though, constructing and using a system to predict currency and banking crises is not easy, mechanically applied systems do not have a good track record. The report examines the reasons existing mechanisms have performed poorly and develops a benchmark early warning system for Korea, based on widely accessible data from a variety of Asian countries. The data are examined by two different tools: first, the behavior of a large number of variables before, during, and after crises is examined. Second, a standard statistical technique, probit, is used to link crises to a number of variables simultaneously.