This paper provides a comprehensive survey of deep Facial Expression Recognition (FER) systems, focusing on methods, benchmarks, models, and datasets. It begins with a detailed timeline of the evolution of FER methods and datasets, highlighting the transition from traditional machine learning to deep learning. The paper then reviews the basic principles of FER, including preprocessing, feature extraction, and classification, and discusses the challenges of overfitting and expression-unrelated variations. It introduces benchmark datasets categorized into controlled and uncontrolled environments, and evaluates different FER models. The paper also covers popular deep neural networks and training strategies for static and dynamic image sequences, and identifies remaining challenges and future directions for robust deep FER systems. Key contributions include a new taxonomy of FER methods and datasets, a comprehensive review of cutting-edge results, and suggestions for future research.This paper provides a comprehensive survey of deep Facial Expression Recognition (FER) systems, focusing on methods, benchmarks, models, and datasets. It begins with a detailed timeline of the evolution of FER methods and datasets, highlighting the transition from traditional machine learning to deep learning. The paper then reviews the basic principles of FER, including preprocessing, feature extraction, and classification, and discusses the challenges of overfitting and expression-unrelated variations. It introduces benchmark datasets categorized into controlled and uncontrolled environments, and evaluates different FER models. The paper also covers popular deep neural networks and training strategies for static and dynamic image sequences, and identifies remaining challenges and future directions for robust deep FER systems. Key contributions include a new taxonomy of FER methods and datasets, a comprehensive review of cutting-edge results, and suggestions for future research.