This paper deals with the extraction of facial features from images presenting frontal views of human faces. The proposed technique combines color and structural characteristics of facial features and human faces. The technique relies on an initial generic segmentation based on color information. Regions allow a more robust estimation of the facial features location. Mouth region candidates are first selected based on color and shape information. From these mouth candidates, and applying geometrical information, eyes and eyebrows region candidates are defined. All sets of candidates are jointly analyzed and the most likely set, from a structural viewpoint, is selected. The technique has been tested with a large set of images, both from fixed frontal view databases (XM2VTS) as well as from generic ones (MPEG4 and MPEG7 databases).