Abstract Studies of high‐frequency (above 1 Hz) earthquake source processes are important not only to clarify the earthquake source process on smaller length scales but also to quantitatively predict strong ground motion. However, the application of conventional waveform inversion methods is not straightforward for high frequencies, because random heterogeneities in the Earth cause incoherent scattered waves and the source process is also hard to treat deterministically. To obviate these difficulties, seismogram envelope inversion methods have been developed since the 1990s for clarifying high‐frequency earthquake source processes. In this chapter, we first give a broad discussion of the methods in terms of data types, Green's function, source parameters, inversion methods, and so on. We developed an envelope inversion method in 1998, in which we used theoretical envelope Green's functions based on the radiative transfer theory as a propagator from a source to a receiver, and estimated the spatial distribution of high‐frequency seismic energy radiation from an earthquake fault plane. We have applied the envelope inversion method to nine moderate‐to‐large earthquakes. Here, we compile the results and clarify some characteristics of high‐frequency seismic energy radiation from moderate‐to‐large earthquakes. Concerning a scaling of high‐frequency radiated energy, logarithm of the high‐frequency seismic energy is found to be proportional to the moment magnitude with a coefficient of proportionality of 1, which is different from 1.5 for whole‐band seismic energy. Moreover, a regional difference in high‐frequency seismic energy radiation is detected for the earthquakes analyzed: Earthquakes in offshore regions of northeastern Japan are found to be more energetic by about an order of magnitude than inland earthquakes in Japan and Taiwan. Regarding the spatial relations, we find four earthquakes in which high‐frequency radiation occurs dominantly at the edges of asperities (areas of large fault slip); in four cases there is no correlation between locations of high‐frequency radiation and asperities. For one earthquake, we have no fault slip model. So far, reasons for the variation are not known yet, heterogeneous distribution of stress, strength, and material properties may control the variability. These characteristics will provide important information for the study of high‐frequency earthquake source process and improvements for predicting strong ground motion.