Estimates of the radiated seismic energy based on teleseismic and regional data often differ by up to an order of magnitude, with a tendency for regional estimates to be larger than teleseismic estimates for the same event. In this study we compare the velocity spectrum determined from teleseismic data after correction for radiation pattern and propagation effects, with the velocity spectrum determined from regional data, after the corresponding corrections, for nine earthquakes in the Middle America subduction zone of Mexico. This comparison of the corrected spectra is used to identify and reduce the sources of the regional versus teleseismic energy discrepancy, which is about an order of magnitude for these events. We find that the teleseismic attenuation operator needs to be calibrated. In our case, for the tectonic environment of the Mexican subduction zone, we need a teleseismic attenuation operator that is stronger at high frequencies than the global average. A larger factor, however, is the correction needed to account for site amplification. This correction has an impact on both regional and teleseismic data, but it has a larger influence on the regional estimates because the angle of incidence for teleseismic waves is steep and the stations are located on more competent rock. By modifying the teleseismic attenuation operator and applying site corrections based on a generic site model, we essentially eliminate the order-of-magnitude discrepancy between teleseismic and regional estimates of the radiated seismic energy for these events.