To assess the predictive content of the interest rate term spread for future economic growth, we distinguish short-run from long-run predictability by using two different approaches. First, following Dufour and Renault (1998) a test procedure is proposed to test for causality at different forecast horizons. Second, the framework of Geweke (1982) and Hosaya (1991) is used to construct a simple test for causality in the frequency domain. This methodology is applied to investigate the predictive content of the yield spread for future output growth. For U.S. data we observe good leading indicator properties at frequencies around one year and typical business cycle frequencies. Using German data we found a (rather weak) predictability at low frequencies only.