The aim of the present article is to further check the use of the horizontal-to-vertical (h/v) spectral ratio, which has been recently suggested as an indicator of site effects. The data set consists of 110, three-component, high sensitivity accelerograms, recorded at five different depths by the Garner Valley Downhole Array (GVDA), in southern California, with peak ground accelerations 0.0002 g ≦ ag ≦ 0.04 g, magnitudes 3.0 ≦ ML ≦ 4.6, and hypocentral distances 16 km ≦ R ≦ 107 km. First, the stability of the (h/v) spectral ratio is investigated by computing the mean for the whole data set in different depths. The (h/v) spectral ratio on the surface is compared with the surface-to-depth standard spectral ratio, with theoretical S-wave transfer functions derived from the vertical geotechnical profile, as well as with the (h/v) spectral ratio of synthetic accelerograms generated by the discrete wavenumber method. Both theoretical and experimental data show a good stability of the (h/v) spectral ratio shape, which is in good agreement with the local geological structure and is insensitive to the source location and mechanism. However, the absolute level of the (h/v) spectral ratio depends on the wave field and is different from the surface-to-depth spectral ratio. Consequently the (h/v) spectral ratio technique provides only partially the information that can be obtained from a downhole array. But surface-to-depth ratios may also be misleading because they combine effects at surface and at depth.