We find that the composite leading index (CLI) is useful for forecasting gross national product (GNP), both in sample and in an out-of-sample real-time exercise. We propose a nonlinear specification in which cyclical shifts of the CLI precede those in GNP. However, we find that better forecasts are provided by a simple linear relation between current GNP growth and the growth rate of the CLI during the previous quarter along with an error-correction term corresponding to the previous quarter's logarithmic difference between the level of the CLI and the level of GNP.