Molecular representations are central to computational chemistry, particularly in the application of deep learning for molecular design and predictive modeling. Representing molecular structures involves capturing various levels of chemical information, from elemental composition and bonding in one-dimensional or two-dimensional forms to the spatial configurations critical to three-dimensional (3D) molecular systems. These 3D representations are essential for tasks such as predicting molecular interactions and designing drugs. Three primary types of 3D molecular representations—Cartesian coordinates, distance matrices, and voxel-based grids—are widely used in deep learning applications. These formats facilitate the accurate prediction and generation of molecular structures by reflecting real-world spatial properties. However, challenges remain in integrating 3D information into machine learning models, including achieving rotational and translational invariance. Recent advancements in deep learning architectures highlight the growing potential of 3D molecular representations in improving molecular generation, accelerating drug discovery, and advancing materials science. These representations continue to drive innovation in computational chemistry by enabling more precise modeling of molecular systems.