The detection of the 2D position of features is dependent upon the facial orientation, so for some applications it is better to detect 3D features. 3D positions can be extracted by stereoscopic measurements with the use of 2 images (front view and side view). After extraction, 3D positions are converted to 3D feature vectors defined on an individual head. The number of vectors can be reduced by defining a smaller number of principal components which are taken in combination to form the original vectors. To evaluate the similarities between 2 principal components, a membership function is introduced for every principal component. A face is recognized as the person whose sum of membership values is greatest.