Sometimes, the best available information about an uncertain future is a single forecast. On the other hand, stochastic-programming models need future data in the form of scenario trees. While a single forecast does not provide enough information to construct a scenario tree, a forecast combined with historical data does—but none of the standard scenario-generation methods is suited to handle this combination. In this paper, we present a new scenario-generation method that combines a single forecast with historical forecast errors. The method is purely data driven and can take into account dependencies between errors of forecasts of different lengths.