The subject of this article is to see if the response to S waves at a specific site can be determined from coda waves. We used aftershock records of the 1992 Landers and Big Bear earthquakes observed at four TERRAscope stations (PAS, PFO, SVD, and GSC) in southern California. The wide dynamic range of the TERRAscope instrument provided a broadband seismogram for S waves, and, at the same time, coda waves over a long lapse-time window. We analyzed a total of 708 three-component waveforms from 65 events whose magnitudes ranged from 3.3 to 5.4. The station SVD is located on Quaternary alluvium, and the remaining stations are on rocks of Cretaceous age. We first evaluated the coda site-amplification factor using a spectral ratio method (May-eda et al., 1991; Koyanagi et al., 1992) for frequency bands centered at 0.75, 1.0, 1.5, 2.0, 3.0, 4.0, and 6.0 Hz for lapse-time windows after 70 sec from the origin time. We found that the site-amplification factor for the radial component is similar to that for the transverse component. This similarity was expected because codas consist of backscattered waves coming from all directions into the station, resulting in a directionally averaged site response. We then evaluated the S-wave site-amplification factor using an inversion method (Andrews, 1986) on the direct S-wave portion of the same data set. We found that the coda site response agrees with the S-wave site response within a factor of about 1.5, supporting the earlier findings by Tsujiura (1978) and justifying the use of coda waves in determining site-specific amplification for S waves. This agreement between S-wave site response and coda site response is understandable if we assume that coda waves are dominated by S waves, as theoretically predicted by Aki (1992) and Zeng (1993).