We present a comparison between observed and predicted amplification on a topographic feature. Our data come from an ad hoc experiment carried out during the summer of 1989, in Epire (Northern Greece). Ten digital, autonomous, three-component stations recorded a total of 68 small earthquakes. As these data were inadequate to compute empirical transfer functions using the standard spectral ratio (SSR) technique, we explored the possible use of horizontal-to-vertical spectral ratios (HVSR). First, we compare HVSR computed from theoretical transfer functions for incident P, SV, and Rayleigh waves with the horizontal transfer function for each incidence type. Second, we compute HVSR for our weak-motion data set and show that the results are stable, with a scatter not larger than that usually accepted in SSR studies. We observe significant differences between N-S and E-W motion. When we compare HVSR with theoretical transfer functions, we obtain a better agreement for SV motion than for SH motion. Amplification levels are similar between HVSR and predicted amplification, even for the SH case. This suggests that discrepancies that have been signaled for the topographic amplification between observations and predictions could be due to site effects being present in the reference station used to computed SSR. Our results support the use of HVSR to determine topographic site effects. Thus, HVSR may have a large potential to evaluate site effects in a wider variety of geological conditions than previously thought, without the shortcomings of SSR. Finally, we show that HVSR computed from noise recordings gives a reasonable estimate of HVSR computed from weak-motion data. Although this conclusion should be better substantiated, our preliminary results are encouraging.