Energetic explosions in the atmosphere produce blast waves, which propagate as infrasound (low-frequency sound waves) and can be detectable at long ranges. Numerous empirical models have been developed to relate explosion energies of events (in terms of trinitrotoluene-equivalent yield) to the metrics of observed infrasonic signals (e.g., peak amplitude, period, and impulse per unit area). However, the existing empirical models often do not take into account variability of atmosphere (e.g., temperature, pressure, and wind) and an empirical model determined in a certain meteorological specification may not be valid in another weather scenario. In this case, unmodeled propagation effects can compromise the accuracy of source energy estimates. We propose another approach to determine explosion energy based on numerical simulations. Infrasound propagation in a given meteorological circumstance is simulated by a full 3-D finite difference method providing accurate Green's functions between the source and receivers. With the numerical Green's functions full-waveform inversion of infrasonic signals is performed to obtain the source time history of the event, and the metrics of the acoustic source time function are related to the energy of explosion source. The technique is applied to controlled chemical explosions and volcanic explosions for verification.