In this paper we propose a new approach to automatic human face detection which employs anisotropic Gaussian filters as local image descriptors. We then show how the paradigm of classifier combination can be used for building a face detector that outperforms the current state–of– the–art systems, while remaining fast enough for being used in real–time systems. We report a number of results on some reference datasets and we use an unbiased method for comparing the detectors.