Abstract Facial Emotion Recognition (FER) is a growing field in computer vision that enables machines to detect and classify human emotions through facial expressions. With the advent of deep learning, particularly Convolutional Neural Networks (CNNs), the performance of FER systems has significantly improved, surpassing traditional handcrafted methods. This paper presents a customized ResNet-18 architecture tailored for FER tasks. The proposed model is evaluated on the FER-2013 using various performance metrics such as accuracy, precision, recall, and F1-score. Our results demonstrate that ResNet-18 provides a strong balance between accuracy and computational efficiency.