This paper examines the determinants of currency crises with a panel annual dataset for 30 countries between 1975 and 1996. We estimate a probit model with random effects, and find that high rates of seignorage, current account imbalances, real exchange rate misalignment, low foreign exchange reserves, negative terms of trade shocks, poor growth performance, and a measure of regional contagion all have significant power to explain the presence of currency crises in our sample. In general, our results can be interpreted as supporting both first and second-generation models of currency crises. Various robustness tests confirm the validity of these results. We also find that currency crises have an important predictable component. Using our benchmark regression we are able to predict correctly a majority of the currency crises that occurred within our sample.