This survey reviews the methodologies for data-driven grasp synthesis, focusing on sampling and ranking candidate grasps. The approaches are categorized into three groups based on the type of object: known, familiar, and unknown. Known objects are those for which grasp databases exist, familiar objects are similar to previously encountered ones, and unknown objects require new feature extraction for grasping. The survey discusses common object representations and perceptual processes that facilitate these techniques. Analytic approaches, which rely on geometric, kinematic, or dynamic formulations, are contrasted with empirical or data-driven methods that use sampling and ranking based on existing grasp experience. The paper also highlights the limitations of analytic approaches, such as assumptions about object models and the lack of robustness to positioning errors. Data-driven methods, which are more flexible and adaptable, are detailed, including their use of machine learning, 3D sensing, and sensory data to generate and rank grasp candidates. The survey concludes by discussing open problems in robot grasping and drawing parallels with classical analytic approaches.This survey reviews the methodologies for data-driven grasp synthesis, focusing on sampling and ranking candidate grasps. The approaches are categorized into three groups based on the type of object: known, familiar, and unknown. Known objects are those for which grasp databases exist, familiar objects are similar to previously encountered ones, and unknown objects require new feature extraction for grasping. The survey discusses common object representations and perceptual processes that facilitate these techniques. Analytic approaches, which rely on geometric, kinematic, or dynamic formulations, are contrasted with empirical or data-driven methods that use sampling and ranking based on existing grasp experience. The paper also highlights the limitations of analytic approaches, such as assumptions about object models and the lack of robustness to positioning errors. Data-driven methods, which are more flexible and adaptable, are detailed, including their use of machine learning, 3D sensing, and sensory data to generate and rank grasp candidates. The survey concludes by discussing open problems in robot grasping and drawing parallels with classical analytic approaches.