This chapter deals with the analysis of multitrial electrophysiology datasets coming from neuroelectromagnetic recordings by electro-encephalography and magneto-encephalography (EEG and MEG). For such measurements, multitrial recordings are necessary in order to extract meaningful information. The obtained datasets present several characteristics: no ground-truth data, high level of noise (defined as the part of the data which is uncorrelated across trials), inter-trial variability. This chapter presents tools that deal with such datasets and their properties. The focus is on two families of data processing methods: data-driven methods, in a section on non-linear dimensionality reduction, and model-driven methods, in a section on Matching Pursuit and its extensions. The importance of correctly capturing the inter-trial variability is underlined in the last section which presents four case-studies in clinical and cognitive neuroscience.