This paper provides a comprehensive review of deep learning for anomaly detection, a critical and active research area. It discusses the unique complexities and challenges in anomaly detection, such as unknownness, heterogeneous anomaly classes, rarity, and class imbalance. The paper categorizes deep anomaly detection methods into three main categories: deep learning for feature extraction, learning feature representations of normality, and end-to-end anomaly score learning. Each category is further divided into subcategories based on different modeling perspectives. The paper reviews the key intuitions, objective functions, underlying assumptions, advantages, and disadvantages of these methods, and discusses how they address the challenges in anomaly detection. It also explores future opportunities and new perspectives to address these challenges. The review includes a detailed discussion on various deep learning techniques, such as autoencoders, generative adversarial networks (GANs), and predictability modeling, and their applications in different types of data, including high-dimensional, temporal, and graph data. The paper aims to provide a deep understanding of the problem nature and the state-of-the-art in deep anomaly detection, offering insights into the strengths and limitations of current methods and suggesting directions for future research.This paper provides a comprehensive review of deep learning for anomaly detection, a critical and active research area. It discusses the unique complexities and challenges in anomaly detection, such as unknownness, heterogeneous anomaly classes, rarity, and class imbalance. The paper categorizes deep anomaly detection methods into three main categories: deep learning for feature extraction, learning feature representations of normality, and end-to-end anomaly score learning. Each category is further divided into subcategories based on different modeling perspectives. The paper reviews the key intuitions, objective functions, underlying assumptions, advantages, and disadvantages of these methods, and discusses how they address the challenges in anomaly detection. It also explores future opportunities and new perspectives to address these challenges. The review includes a detailed discussion on various deep learning techniques, such as autoencoders, generative adversarial networks (GANs), and predictability modeling, and their applications in different types of data, including high-dimensional, temporal, and graph data. The paper aims to provide a deep understanding of the problem nature and the state-of-the-art in deep anomaly detection, offering insights into the strengths and limitations of current methods and suggesting directions for future research.