Pose variations represent, along with illumination variations, a very important issue in creating a robust face recognition system. In order to cope with different face orientations, which are inherent in real-life conditions, one approach is to store in the database for each individual multiple images with different poses. However it is difficult to cover the entire domain of pose variations to make the system completely invariant to new poses. Our approach is based on normalizing the pose of every face to have the same orientation. In order to achieve that we are using a statistical method of modelling the face appearance called active appearance models (AAMs).