The article attempts to combine Poland’s macroeconomic forecasts in 1995-2002 covering three categories: inflation, GDP growth and unemployment. A combined forecast makes use of information offered by several different forecasts, so it makes it possible to reduce the forecast error. Both “simple forecast combination methods” (arithmetic mean and median) and “complex forecast combination methods” (variants of the classic variance-covariance method and the Bayes method) were used. The results of the research show that the average error of simple combined forecasts, though worse than that of the best individual forecast, is always lower than the average error of individual forecasts. On the other hand, in the case of complex methods of combining forecasts, the average errors of these forecasts are usually greater than both the average error of individual forecasts and the average error of simple methods of combining forecasts. The work explains the causes for this situation and offers suggestions on future research into combining Polish macroeconomic forecasts.