The aim of this work is to check the stability of the horizontal-to-vertical spectral ratios (hvsrs) calculated at the Venosa station site (Italy). This site lies over a layer of anthropogenic fill (4 m thick), a rigid layer of conglomerates (15 m thick), and a thick layer of clays (about 300 m thick) above the seismic bedrock. The velocity inversion, which takes place at the conglomerates–clays interface, is of main importance for the amplification behavior of this site. We have analyzed nearly 2 years of data, composed of 244 triggered noise records and 44 earthquakes. The results obtained by the two data sets show different site-response characteristics. In particular, the earthquake hvsr is not deamplified in the frequency range 1–8 Hz like the triggered noise hvsr. To find out the origin of this difference, we modeled both the triggered noise and the earthquakes, taking advantage of an improved version of the Thompson–Haskell propagation matrix method. The differences between triggered-noise- and earthquake-amplification functions might be explained by the difference in composition and propagation of the seismic wave fields. Moreover, we show that the nonlinear behavior of the anthropogenic fill might explain the presence of the misfit of the resonance frequency attributed to this layer between triggered noise and earthquakes.