In the field of facial recognition, so called embeddings are used in state of the art face recognition models. These embeddings are a mapping of a picture of a face into a vector. These vectors can be easily compare to each other through a distance measurement. Instead of embedding the complete face this thesis explores the possibilities of embedding individual facial features separately and training neural networks with them. Each selected facial feature gets trained on a separate neural network and gets their own embedding. The results in facial recognition accuracy can be compared each other and analysed. For finding facial similarities a user study is conducted that tests the capability of the proposed model. The results of this thesis can be used as a basis for future work and shows what can be improved.