After the base forecasts are generated via physical forecasting methods, the next logical step is to post-process those initial forecasts using statistical and machine-learning methods, for the physics-based forecasts are almost always associated with model-led bias and under-dispersion, which must be reduced and/or calibrated if not trimmed completely. As the raw forecasts can be either deterministic or probabilistic, post-processing can be viewed as a conversion step, which results in four distinct directions: (1) deterministic-to-deterministic, (2) deterministic-to-probabilistic, (3) probabilistic-to-probabilistic, and (4) probabilistic-to-probabilistic conversion. This chapter reviews the different mechanisms with which post-processing can be conducted, as a means to improve the quality of the raw forecasts.