This paper provides a comprehensive review of deep learning methods for anomaly detection, discussing their key intuitions, objectives, assumptions, and capabilities in addressing challenges in anomaly detection. Anomaly detection, which identifies data instances that deviate from the majority, has been an active research area for decades. Recent advances in deep learning have enabled more effective anomaly detection by learning feature representations or anomaly scores via neural networks. The paper categorizes deep anomaly detection into three main frameworks: deep learning for generic feature extraction, learning representations of normality, and end-to-end anomaly score learning. It also presents a hierarchical taxonomy of 11 fine-grained categories of methods. The paper discusses the unique problem complexities and challenges in anomaly detection, including low recall rates, high-dimensional data, data-efficient learning of normality/abnormality, noise resilience, complex anomalies, and anomaly explanation. Deep learning methods offer end-to-end optimization of the anomaly detection pipeline and the ability to learn representations tailored for anomaly detection, which are crucial for addressing these challenges. The paper also highlights the advantages and disadvantages of deep learning methods compared to traditional approaches, and discusses future opportunities for research in this area.This paper provides a comprehensive review of deep learning methods for anomaly detection, discussing their key intuitions, objectives, assumptions, and capabilities in addressing challenges in anomaly detection. Anomaly detection, which identifies data instances that deviate from the majority, has been an active research area for decades. Recent advances in deep learning have enabled more effective anomaly detection by learning feature representations or anomaly scores via neural networks. The paper categorizes deep anomaly detection into three main frameworks: deep learning for generic feature extraction, learning representations of normality, and end-to-end anomaly score learning. It also presents a hierarchical taxonomy of 11 fine-grained categories of methods. The paper discusses the unique problem complexities and challenges in anomaly detection, including low recall rates, high-dimensional data, data-efficient learning of normality/abnormality, noise resilience, complex anomalies, and anomaly explanation. Deep learning methods offer end-to-end optimization of the anomaly detection pipeline and the ability to learn representations tailored for anomaly detection, which are crucial for addressing these challenges. The paper also highlights the advantages and disadvantages of deep learning methods compared to traditional approaches, and discusses future opportunities for research in this area.