Analyses and checks the annual forecasts produced each autumn from four prominent UK economic modelling organizations. Compares these forecasts with those of three Bayesian vector-autoregressive models. Examines the accuracy for each set of forecasts up to four years ahead and for different horizons. Examines the direction of the forecasts and the effect of forming simple combinations of the different forecasts. Finds evidence that while the BVAR forecasts are inferior to those from the economic models, they contain information which could be used in order to improve the other forecasts.