Over the last decades, face detection and facial features detection have received a great deal of attention from the scientific community, since these tasks are essential for a number of important applications, such as face recognition, face tracking, human-computer interaction, face expression recognition, security, etc. This work proposes the use of a classifier based on FERNS to recognize interest points across images and then detect and track the facial features. We begin with a brief review of the most common approaches used in facial features detection and also the theory around the FERNS. In addition, an implementation of the facial features detection based on FERNS is present to provide results and conclusions. The method proposed here relies on an offline training phase during which multiple views of the keypoints to be matched are synthesized and used to train the FERNS. The facial features detection is performed on images acquired in real-time from many different viewpoints and under different lighting conditions.