The combination of individual forecasts is often a useful tool to improve forecast accuracy. There is a large number of alternatives to combine forecasts. The most commonly used is to assign equal weights for all the individual forecasts (mean). This has proven to be hard to beat by most sophisticated alternatives, like basing the weighting mechanism in the variance-covariance matrix of the forecast errors or ridge regressions, among others. As the combination of forecasts can be seen as a dimension reduction problem (from J expert forecasts to just a single one), we consider the use of reduction dimension techniques to produce this single forecast. Since the goal is forecasting, it seems that Partial Least Squares (PLS from now on) that takes into account the covariance of each �x� (forecast in our case) with �y� (the variable to forecast) should give better results than other alternatives that do not consider the information in the variable that we want to forecast when reducing the dimension. The data used are the forecasts provided by the Survey of Professional Forecasters for the main US macroeconomic aggregates.