Abstract Microzonation studies using ambient noise measurements have become very popular in the last years for local soil characteristics determination. For defining the capabilities of the ambient noise based methods, a physical understanding of the nature of noise is essential. In this way, many studies have shown that the noise wavefield composition is heterogeneous, comprising both body and surface waves with fundamental and upper modes. In this work, a recently developed method of time-frequency polarization analysis, based on the stationary wavelet packet transform, is used to study how the waves’ polarization influences in the application of the HVSR method. For this purpose, each recorded signal is first decomposed into a set of sub-signals, according to their reciprocal ellipticity. After that, the H/V analysis is carried out on the raw recorded signal, as well as on the obtained sub-signals, evaluating the contribution of each sub-signal (i.e., each reciprocal ellipticity range) to the expected H/V result. For the study, single station seismic noise was recorded at 11 sites around the province of Alicante (southeastern Spain). The obtained results illustrate that high frequency peaks, above 1–2 Hz, are clearly represented by just using the sub-signals associated to the ellipticity range of [0–0.1], whereas low frequency peaks, down 0.5 Hz, need a wider ellipticity interval for being properly retrieved, i.e., [0–0.4] or even [0–0.6]. Therefore, linearly polarized waves’ contribution, range [0–0.1], is enough for retrieving the high frequency H/V peaks, but not for obtaining the complete H/V curve, especially the low frequency peaks. In these cases, the elliptical polarized Rayleigh waves’ contribution cannot be misestimated in HVSR computation.