AbstractThe analysis and estimation of extreme event occurrences is a central problem in many fields of geoscience. Advancements in the study of extreme events have recently been limited, arguably in connection with asymptotic assumptions in the traditional extreme value theory (EVT) and with its focusing on a small fraction of the available observations representing the tail properties of the underlying event generation process. Here we develop a Metastatistical Extreme Value framework (MEV) which relaxes limiting assumptions at the basis of the traditional EVT and accounts for the full distribution of the underlying “ordinary” events. We apply this general approach to the relevant case of daily rainfall and find that the MEV approach reduces the uncertainty in the estimation of high‐quantile extremes by up to 50% with respect to the classical EVT. The improved predictive power of the MEV framework is connected with its recognizing that extremes emerge from repeated sampling of ordinary events, thereby being able to use all available observations.