An application of neural networks for the classification of astronomical images is presented. A neural network classifier feeded with basic image parameters is shown to achieve extremely good results in star/galaxy separation. The applicability to very large digital surveys such as DENIS or 2MASS is discussed.