Physical justification is provided for the use of kinetic energy backscatter in forecast models, particularly in respect of ensemble prediction systems. The rate of energy backscatter to scales near the truncation limit is controlled by a total energy dissipation function involving contributions from numerical diffusion, mountain drag and deep convection. A cellular automaton is used to generate evolving patterns that, together with the dissipation function, define a stream-function forcing field. Each member of the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecast system is perturbed by a different realization of this backscatter forcing, and the resulting increase in ensemble spread, if not excessive, has a beneficial impact on probabilistic measures of forecast skill. The input of small-scale kinetic energy by the backscatter algorithm also helps to correct a known problem with the energy spectrum in the ECMWF model—the absence of the observed −5/3 spectral slope in the mesoscales. Copyright © 2005 Royal Meteorological Society.