Simulation-based inference (SBI) is a suite of methods in which the model is explicated not by an explicit numerical joint probability density but through a simulator: a forward process from causes to effects. This chapter presents an overview of neural SBI methods (summarisation, (joint/likelihood/posterior) density estimation, and (likehood/likelihood-to-evidence) ratio estimation), elucidating the similarities in their formulations. It then discusses strategies to improve and accelerate training, among which prior truncation. Finally, it presents the framework for efficiently diagnosing, validating, and calibrating SBI results through amortisation: the ability to nearly instantaneously analyse a large number of simulated data sets.