This chapter provides an overview of machine learning force fields, a rapidly evolving area that integrates physics-based modeling and machine learning. Force fields are models for atomic interactions, often based on fixed functional forms and empirically derived parameters. While force fields have become essential tools for many fields including structural biology, drug design, and materials discovery, they have limitations in modeling complex biological systems. Machine learning force fields (MLFF) offer a promising data-driven approach that is more flexible and more accurate than traditional force fields. We begin with a survey of the history and theoretical background of classical force fields, followed by the introduction of the key aspects for constructing MLFF, such as molecular representation, model architecture, and target properties. Different types of MLFFs, including multilayer perceptrons, graph neural networks, and kernel-based methods, are discussed, highlighting their advantages and disadvantages in different scenarios. The chapter also summarizes popular datasets used for training MLFFs, which might be helpful not only for developing new MLFFs but also for improving other chemical models. We conclude by addressing the challenges and future directions of the field.