This chapter introduces the Sampling-and-Learning (SAL) framework, a unifying approach to understanding derivative-free optimization (DFO) algorithms. The SAL framework consists of two main stages: sampling, where new candidate solutions are generated, and learning, where promising solutions are selected to guide future sampling. The framework iteratively alternates between these stages, using a learned model to represent the algorithm’s belief about promising regions in the solution space. The chapter also presents a simplified version called the Sampling-and-Classification (SAC) framework, which uses binary classification to distinguish between promising and unpromising solutions. The SAL and SAC frameworks provide a statistical perspective on how DFO algorithms balance exploration and exploitation, offering a systematic way to analyze and design optimization methods. The chapter concludes by discussing the challenges and potential of these frameworks, including the computational overhead of learning and the need for accurate models to guide the search effectively.