This paper describes an empirical study of over 4000 specific share return forecasts made by 35 UK stockbrokers and by the internal analysts of a large UK investment institution. A comparison of forecast and realised returns reveals a small but potentially useful degree of forecasting ability. A large part of the information content of the forecasts, however, appears to be discounted in the market place within the first month. Nevertheless, an analysis of some 3000 transactions motivated by, and executed at the time of, the forecasts shows that the apparent predictive ability of the recommendations could be translated into superior performance by the fund's investment managers. Differences in forecasting ability between brokers do not appear to persist over time, but predictive accuracy can be improved by pooling simultaneous forecasts from different sources.