Face recognition technology is increasingly used in security, healthcare, and law enforcement. This study evaluates the performance of three deep learning architectures: VGG16, AlexNet, and a lightweight convolutional neural network (LWCNN) on two facial datasets: Labeled Faces in the Wild (LFW) and the Olivetti Research Laboratory (ORL) dataset. We aim to test the hypothesis that a lightweight CNN can provide comparable performance while reducing computational demands compared to larger models like VGG16 and AlexNet. Our results show that all models perform well on both datasets. On the LFW dataset, VGG16 achieved 97.50% accuracy, LWCNN 91.67%, and AlexNet 83.33%. On the ORL dataset, VGG16 achieved 100.00%, while AlexNet and LWCNN reached 97.50% and 96.00% respectively. These results suggest that while more complex models like VGG16 and AlexNet deliver higher performance, the lightweight CNN provides a competitive alternative, particularly for resource-constrained environments. This work contributes to the development of efficient face recognition systems by highlighting the trade-offs between model complexity and performance, especially in real-world applications involving occlusion and varying conditions.