By means of three-dimensional (3-D) numerical simulations, including the Laboratoire Souterrain à Bas-Bruit (LSBB) topography, we carefully analyze site effects assessments yielded by two approaches: the classical site to reference spectral-ratio method (SRM) and the statistical median reference method (MRM). We show for both isotropic and double-couple point sources that a 94% reduction in the number of stations of a regularly spaced array yields MRM site-effect estimates within 5% of those obtained from the absolute regional median, and within 20% using a 98% station reduction with irregularly located sites. In contrast, the SRM yielded site-effect overestimates greater than 50% in some areas and up to 100% in specific sites, which makes the MRM much more robust than the SRM. We determined a 33% probability to exceed an amplification factor of 2, and an 8% probability to exceed a factor of 3 due to topography in the surroundings of the sharpest summit of the LSBB area.