On 17 August 1999, the MW 7.4 Izmit earthquake occurred in northwestern Turkey. A temporary seismic network was installed to improve the geometry of a previously installed network. In this study, 262 aftershocks of the Izmit earthquake with magnitude ML ranging between 0.4 and 4.5 were analyzed using digital recordings from 17 stations of these networks. S-wave and P-wave spectral records, corrected for path attenuation function, were inverted by means of the generalized inversion technique (GIT). The GIT site responses were compared with those calculated by horizontal-to-vertical (H/V) ratios applied to both earthquake data and to pre-event noise. In most cases, the H/V of the earthquake data provided site responses with shapes consistent with those obtained using GIT. However, the level of amplification was occasionally found to be different. In general, these discrepancies can be explained either by the amplification affecting the vertical component of the ground motion that we detected by GIT or by the waves having a propagation in almost the vertical direction. Using the H/V ratio applied to noise, for most cases, we obtained not only the peaks for the fundamental resonance frequencies, but site responses with shapes similar to those obtained by GIT. The detected level of amplifications using the H/V ratio showed a tendency not to exceed those calculated by the inversion. Finally, amplifications up to a factor 5 are found for stations placed over complex topography. For these sites, amplifications are not always correctly identified by the H/V technique, using either earthquakes or noise.