This paper explores the use of a maximum entropy econometric approach to combine forecasts when the small amount of information available does not allow the use of regression procedures since a dimensionality problem arises. This approach has its roots in information theory and builds on the entropy information measures and the classical maximum entropy principle, which was developed to recover information from underdetermined models. More specifically, we use the maximum entropy econometric approach for the measure of Shannon and we also propose its extension to the quadratic uncertainty measure. The experimental results over a pool of forecasts referring to Spanish inflation show some improvements when compared with equally weighted combined forecasting. Copyright © 2011 John Wiley & Sons, Ltd.