Multiclass semantic image segmentation is widely used in a variety of computer vision tasks, such as object segmentation and complex scene understanding. As it decomposes an image into semantically relevant regions, it can be applied in segmentation of face images. In this paper, an algorithm based on multiclass semantic segmentation of faces is proposed using conditional random fields. In the proposed model, each node corresponds to a superpixel, while the neighboring superpixels are connected to nodes through edges. Unlike previous approaches, which rely on three or four classes, the label set is extended here to six classes, i.e. hair, eyes, nose, mouth, skin, and background. The proposed framework is evaluated on standard face databases FASSEG, FIGARO, and LFW. Experimental results reveal that the performance of the proposed model is comparable with state-of-the-art techniques on these standard databases.