This review article provides an overview of deep learning methods in deterministic computational mechanics. Five main categories are identified: simulation substitution, simulation enhancement, discretizations as neural networks, generative approaches, and deep reinforcement learning. The focus is on methods rather than applications, enabling researchers to explore this field more effectively. The review is aimed at researchers new to the field or seeking an overview of deep learning in computational mechanics. Key concepts are explained simply. Simulation substitution replaces the entire simulation with a surrogate model, typically a deep neural network. This can be done through data-driven modeling, which relies on labeled data, or physics-informed learning, which incorporates physical constraints. Simulation enhancement involves replacing parts of the simulation chain while maintaining classical methods for the rest. Discretizations as neural networks use neural network components to construct discretizations, allowing the use of deep learning techniques like automatic differentiation. Generative approaches create new data based on existing data, useful for design space diversification. Deep reinforcement learning uses agents to learn interactions with environments, with the environment modeled by physical equations. The review highlights the importance of understanding the unique taxonomy of deep learning in computational mechanics, distinguishing it from application-oriented perspectives. It discusses various neural network architectures, including convolutional, graph, and recurrent networks, and their applications in computational mechanics. The article also addresses challenges such as robustness, explainability, and handling complex data. It emphasizes the potential of deep learning in computational mechanics, particularly in surrogate modeling and data-driven approaches, while noting the need for further research to improve optimization and generalization. The review concludes with a discussion of future directions and the importance of integrating physics into deep learning models.This review article provides an overview of deep learning methods in deterministic computational mechanics. Five main categories are identified: simulation substitution, simulation enhancement, discretizations as neural networks, generative approaches, and deep reinforcement learning. The focus is on methods rather than applications, enabling researchers to explore this field more effectively. The review is aimed at researchers new to the field or seeking an overview of deep learning in computational mechanics. Key concepts are explained simply. Simulation substitution replaces the entire simulation with a surrogate model, typically a deep neural network. This can be done through data-driven modeling, which relies on labeled data, or physics-informed learning, which incorporates physical constraints. Simulation enhancement involves replacing parts of the simulation chain while maintaining classical methods for the rest. Discretizations as neural networks use neural network components to construct discretizations, allowing the use of deep learning techniques like automatic differentiation. Generative approaches create new data based on existing data, useful for design space diversification. Deep reinforcement learning uses agents to learn interactions with environments, with the environment modeled by physical equations. The review highlights the importance of understanding the unique taxonomy of deep learning in computational mechanics, distinguishing it from application-oriented perspectives. It discusses various neural network architectures, including convolutional, graph, and recurrent networks, and their applications in computational mechanics. The article also addresses challenges such as robustness, explainability, and handling complex data. It emphasizes the potential of deep learning in computational mechanics, particularly in surrogate modeling and data-driven approaches, while noting the need for further research to improve optimization and generalization. The review concludes with a discussion of future directions and the importance of integrating physics into deep learning models.