The application of the horizontal-to-vertical (H/V) spectral ratio technique to earthquake recordings can lead to significant differences in the estimate of the site response compared to that produced by the reference-site method (RSM). In particular, whereas the estimates of the resonance frequency from the two methods are fairly consistent, the levels of amplification are not. Using numerical modeling based on an improved Thompson–Haskell propagator matrix method, we were able to isolate the contribution of pure and converted waves to the site response. We show that the conversion of body waves at the sediment–bedrock interface leads to differences in the site response estimates obtained by the H/V method and the RSM. Such differences are consistent with observations in the field. In particular, the lower level of amplification obtained by the H/V method at frequencies higher than the fundamental one is due to a transfer of energy onto the vertical component caused by S- to P-wave conversion. Applying the RSM technique to the vertical as well as the horizontal component of the P-wave window, we obtain stable estimates of the fundamental resonance frequency that are consistent with 1D resonance for vertically incident P and S waves, respectively. The amplification from the P-wave window of the horizontal component might be considered as a lower boundary.