The National Hurricane Center in Miami and the Joint Typhoon Warning Center in Honolulu have recently extended their tropical cyclone track and intensity forecasts from 3 to 5 days. As the forecast length increases, the ability to predict the initial formation of storms becomes more important. However, objective tools for prediction of tropical cyclone formation are fairly limited. In this paper, a new method for estimating the probability of storm formation within a 24-hour period is described. The method uses GOES water vapor imagery to identify regions with persistent convection, and NCEP global model analyses to determine the conditions in the environment surrounding the disturbances. This information is used as input to linear discriminant and nonlinear neural network algorithms. The performance of these two algorithms will be compared for the Atlantic and east Pacific tropical cyclone basins. The potential for extending the period of prediction by combining this diagnostic method with global model forecasts will also be discussed.