We have applied the regional S-wave coda calibration technique of Mayeda et al. (2003) to earthquake data in and around the Nevada Test Site (NTS) using four regional broadband stations from the Lawrence Livermore National Laboratory (LLNL) seismic network. We applied the same path and site corrections to tamped nuclear explosion data and averaged the source spectra over the four stations. Narrowband coda amplitudes from the spectra were then regressed against inferred yield based on the regional mb(Pn) magnitude of Denny et al. (1987), along with the yield formulation of Vergino and Mensing (1990). We find the following: (1) the coda-derived spectra show a peak that is dependent upon emplacement depth, not event size; (2) source size estimates are stable for the coda and show a dependence upon the near-source strength and gas porosity; (3) for explosions with the same mb(Pn) or inferred yield, those in weaker material have lower coda amplitudes at 1–3 Hz.