The shear (SH)-wave transfer function and the horizontal-to-vertical (HV) spectral ratio are essential to estimate the S-wave velocity pro- le and thickness of surface layers overlying a bedrock on the basis of resonance frequencies. In practice, it is the second method the most used. In this work, we propose a full-wave numerical method, based on a pseudospectral spatial differentiation, to simulate SH and P-S waves generated by random sources distributed spatially and temporally (ambient noise). The modeling allows us to implement seis- mic attenuation, surface waves and causal source radiation patterns, based on random values of the angles of the moment tensor at each source location. We focus on the location of the resonance peaks, since this property is strictly related to the thickness of the layers. First, we analyze Lamb’s problem for which an analytical P-S solution exists. The modeling algorithm is veri ed for a Ricker time history, but the analy- sis can be performed by using spikes as sources. The experiments based on ambient noise are compared to those of a coherent line source as a reference spectrum (e.g., an earthquake event far away from the receivers). SH-wave resonance frequencies can be identi ed in the spectra only when the random sources are located below the bedrock. In the case of P-S waves, the SH-wave transfer function is a good approximation to the HV spectrum, mainly when the noise is generated in the bedrock. Finally, we have assumed a square basin and found that coherent (e.g., earthquake-type) sources may yield identi able peaks but ambient noise gives unreliable results.