A data set of 50 horizontal strong-motion accelerograms from 18 main events from moderate and large earthquakes (5.2 ≤ Mw ≤ 6.9) and distances (9 km ≤ R ≤ 116 km) occurred at three different tectonic settings (normal, strike-slip, and thrust faulting), were used to estimate stress parameters. Three different definitions of stress release were adopted: the classical Brune, the dynamic-rms (root mean square), and the apparent stress. Corner frequencies, f0, based on the Andrews (1986) methodology were computed and, using these calculations, seismic moments were estimated from independent sources (teleseismic data). The calculated corner frequencies seem to be consistent, based on seismic moments, with previous estimates for the same area. Based on those source parameters, a theoretical ω-square spectrum was derived from describing source radiation in terms of earthquake size. Frequency-dependent site amplifications were included in the simulations, separating the recordings in three different categories (B, C, and D) corresponding to (NEHRP) National Earthquake Hazards Reduction Program site classification. The theoretical and observational spectra are in good agreement, in a frequency range from 0.05 to 25 Hz, except for the cases where steep topography affects the recorded data. A poor fit is also observed in cases where the distance of the recorded data is greater than 100 km and the attenuation model seems to be inadequate to describe the theoretical model. The calculated parameters present a mean value of 55 ± 16 bars for Brune stress, 72 ± 26 bars for dynamic-rms stress, and 17 ± 10 bars for apparent stress. For all determined stresses, the thrust-faulting earthquakes exhibited higher values than the calculated normal-and strike-slip-fault seismic events.