In recent years, a limited amount of work has been done on the medium-term linear composite method of forecasting. One common finding in the existing literature is that the consensus forecast measure is a biased predictor of future exchange rates. A widely accepted point of view in exchange rate forecasting research is that no theoretical model should be able to outperform a simple random walk. In this paper, recent exchange rate data and the Granger-Ramanathan linear estimation method are used to test medium-term forecasts. The currencies considered in this study are the most actively traded in the world and include: euros, Japanese yen, Canadian dollars, British pounds and Swiss francs. All currencies are examined relative to the US dollar. The major finding is that the linear composite model does in fact outperform a random walk model and an average forecast for Japanese yen, British pounds and Swiss francs. This evidence suggests that additional research should be conducted on exchange rate forecasting in general and on the linear composite forecast model in particular.