Eighty-five earthquakes recorded in 1991 at the digital surveillance array of Mt. Etna are used for investigating the attenuation of the medium and the amplitude response of the sites. In order to evaluate the influence of the attenuation on seismic spectra, the quality factor Q is estimated for P and S waves by the single-station spectral ratios method, and spectral amplitude ratios calculated between 2 and 20 Hz. The P-wave quality factor Qp is computed from earthquakes recorded at the whole network, while Qs is obtained by S data recorded at three-component stations only. The results indicate that Qp and Qs increase as a function of the frequency according to the law Q = Q0fn (with n ranging from 0.5 to 1.6) as a general rule. Reliable Qs/Qp ratios are obtained at three, three-component stations only, two of which display values ranging between 1.3 and 1.5, while the third displays anomalous values much lower than unity (≈0.5). Significant variations of Qp are found as a function of the depth (much lower Q values are obtained for focal depths less than 5 km) and of the source-station azimuths. After correction of seismic spectra by the geometrical spreading and the quality factor, spectral ratios are computed again at all stations for singling out the site response effects. The amplitude responses differ by up to 10-fold at the different sites. In a very few cases, larger amplification ratios (up to 2 orders of magnitude) are found to affect some frequency bands. Displacement spectra, corrected by the attenuation and the site responses, are then used for inverse determination of the low-frequency level Θ0, the corner frequency ωc, and the fall-off γ of spectra beyond ωc. The latter typically displays ω−3 behavior. The inverse relationship between seismic moment and corner frequency does not hold for events located at shallow depth since very low Q values may locally mask ωc.