We used strong motion records from the 1976 Friuli earthquake (M 6.4) and 10 of the biggest aftershocks recorded by the National Accelerograph Network of the Electrical Power Company of Italy to estimate the quality factor Q of S waves in this region. The wide distance range of the recordings (10<r<190 km) permits us to analyze the spectral amplitude decay of the records using a nonparametric approach Anderson and Quaas, 1988; Castro et al., 1990; Anderson, 1991>. We obtained attenuation functions for a set of 18 frequencies ranging between 0.4 and 25.0 Hz. The values of Q retrieved from the attenuation functions obtained follow the frequency-dependent relation Q=20.4f. A test of the method was made using a second data set consisting of digital seismograms from the Friuli-Venezia Giulia Seismograph Network. In spite of the different size of the volume sampled by these data (10<r<131 km), the frequency dependence of Q obtained (Q=16.1f0.92) is similar to that obtained with the strong motion data set. The near-surface attenuation was also estimated using the model proposed by Anderson and Hough <1984> and Anderson <1991>. We found that &kgr;0 is smaller for the strong motion stations located on rock compared to stations located on either shallow or soft sediments. To estimate the site response of the strong motion stations, we corrected the spectral records for the attenuation effect and then inverted the corrected records to separate source and site effects using the inversion scheme proposed by Andrews <1986>. To verify the site amplification estimates obtained, we also calculated the transfer function of each site using Nakamura's <1989> method for S wave Lermo and Chavez-Garc¿a, 1993>. In general, the shapes of the site functions obtained with the inversion are consistent with the transfer functions obtained calculating the horizontal to vertical component ratio. ¿ American Geophysical Union 1996