Facial emotion recognition (FER) remains a computationally intensive challenge due to micro-variations in expressions, lighting conditions, and demographic diversity. Traditional manual observation methods suffer from subjectivity and inconsistency, while early automated systems relied on handcrafted features with limited generalizability. Recent advances in deep learning, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have enabled end-to-end learning of hierarchical spatial-temporal features directly from raw image data [1]. This study benchmarks CNN and RNN architectures on the FER-2013 dataset (Kaggle), containing 35,887 grayscale images labeled across seven universal emotions: anger, disgust, fear, happiness, sadness, surprise, and neutral. Key innovations include: Experimental results demonstrate an 88% accuracy for RNNs versus 48% for CNNs, highlighting RNNs’ superiority in capturing contextual emotion evolution. Performance metrics (F1-score: 0.86, recall: 0.82) surpass prior works using static CNNs, validating the efficacy of temporal modeling for FER. Applications span affective computing, mental health diagnostics, and intelligent tutoring systems.