Face recognition involves identifying individuals using their facial features. It is the most natural way of recognizing people. With the increase in smart gadgets, automated processes, and easy access to the internet, there comes the need to authenticate people and ensure data security. Face recognition involves two steps—feature extraction and classification. Although the classifier has received most of the focus in recent research, feature extraction methods are sometimes overlooked though they are just as important in determining the model’s effectiveness. This paper aims to address this research gap by systematically evaluating different feature extraction methods—Convolutional Neural Network (CNN) model, Principal Component Analysis (PCA) technique, Histogram of Oriented Gradients (HOG) technique, Local Binary Patterns Histograms (LBPH) algorithm, and the FaceNet model paired with K-Nearest Neighbors (KNN) classifier on the benchmark AT&T dataset by Olivetti Research Laboratory (ORL) in Cambridge, UK. The evaluation measures employed in this study are accuracy, precision, recall, and the F1-score. The findings indicate that the FaceNet model performs noticeably better than other feature extraction techniques when combined with the KNN classifier, most likely because of the higher caliber of embeddings it generates. This study adds to the body of knowledge by emphasizing how crucial it is to choose the right feature extraction techniques, ultimately aiding in improving facial recognition systems.