Recent earthquakes have triggered renewed interest to understand better earthquake site response. Most of the studies comparing various techniques for estimating site response were based on real data (from earthquakes, nuclear blasts, and seismic noise). A theoretical approach, using synthetic data generated with the pseudospectral method, is used to compare four site-response estimation techniques. The limits of applicability of each method were determined by modeling microtremors and incoming SV waves (with different incidence angles) and analyzing the site amplifications. The first two techniques investigated consist of dividing the spectrum of the horizontal motion at a site by that of a reference site using either incident S waves or microtremors. The latter was unable to reveal either the resonant frequencies or peak amplitudes in any cases. The two other techniques are based on the horizontal-to-vertical (H/V) spectral ratio using S waves or microtremors. These techniques were found to reveal at least the fundamental resonant frequency and amplitude (former method only) within 10% error, in the case of simple geology (flat layers). However, the results show that these techniques are unable to take into account 2D effects such as focusing effects and basin-edge effects and yield unreliable or incorrect results in such cases.