We try to separate source effects, propagation path effects and site effects from observed S-waves in frequency domain. Assuming that an observed seismic spectrum is a combination of the source spectrum, the path effect in connection with the geometrical spreading factor and the anelasticity (Q-value) and the site amplification, we can construct the simultaneous logarithmic equations with the unknown quantities such as the source spectrum, Q-value and the site amplification. To solve these equations with linear inversion methods, we set up the criterion that the site effects have a factor of more than 2 due to the free surface amplification. This method is examined by using two data sets; one obtained from 1983 Japan-Sea earthquake (M=7.7) and its largest aftershock (M=7.1) at 6 stations and the other, its 10 aftershocks (M=4.0-6.1) at 3 stations, with hypocentral distances 80-250km and 70-150km, respectively. Only one station (TUC) is common to two data sets. Obtained Q-values of S-waves are propotional to the 0.6 power of frequency in the range of 0.5-8.0Hz. Site effects at TUC from the two independent data set agree well within each standard deviation. From the obtained source spectra, we determine the flat levels of displacement spectra, those of acceleration spectra and the corner frequencies, and then, estimate the seismic moment, crack radius and stress drop of every event.