This article focuses on how close reference rock sites need to be to the sites of interest when using the standard spectral ratio technique to obtain consistent estimates of site amplifications reflecting behavior of seismic waves at the sites of interest. Taber and Smith (1992) and Chávez-García et al. (1999) obtained very different amplification factors for the Parkway Basin, New Zealand, using reference sites only 2 km apart. We analyze 34 earthquakes recorded by a dense 2D array of 19 soil sites in the basin and 4 rock sites fringing the basin. Spectral ratios are calculated separately relative to Taber and Smith's reference site P01, 2 km from the basin, and to a hypothetical average rock site, PR, formed from the four local rock sites, including reference site P25 of Chávez-García et al. Our spectral ratios of the soil sites relative to PR are remarkably consistent, with a distribution of the dominant peaks at 8.4 ± 2.0 at 1.7 ± 0.3 Hz. In contrast, the spectral ratios relative to P01 are much less consistent, and the dominant peak ratios are widely scattered around two separate centers of 14.5 ± 4.1 at 1.8 ± 0.3 Hz and 14.4 ± 5.5 at 4.7 ± 1.0 Hz. Our results relative to the two references are similar to those obtained by Chávez-García et al. and Taber and Smith, respectively. We find that extreme differences between the two are often associated with the occurrence of troughs in the amplitude spectra at P01, possibly due to topographic deamplification there relative to the Parkway Basin in some but not all earthquakes.