This paper empirically investigates the nature of financial crises in the United States before 1914. It attempts to determine whether crises were statistically similar, predictable, and had a common generating mechanism. Using probit and hazard models and out-of-sample criteria, it is shown there are variables that explain movements in the probability of crises and that the probability of crises is seasonal. Two crises were predictable but in the other six episodes every forecasting model examined failed. These results suggest that financial crises were not all statistically alike and that their generation mechanisms differed. Copyright 1994 by Ohio State University Press.