Ground motions at two Kiban-Kyoshin Network (KiK-net) strong motion downhole array sites in Hokkaido, Japan (TKCH08 in Taiki and TKCH05 in Honbetsu) illustrate the importance of three-dimensional (3D) site effects. These sites recorded the M8.0 2003 Tokachi-Oki earthquake, with recorded accelerations above 0.4 g at both sites as well as numerous ground motions from smaller events. Weak ground motions indicate that site TKCH08 is well modeled with the assumption of plane SH waves traveling through a 1D medium (SH1D), while TKCH05 is characteristic of a poor fit to the SH1D theoretical response. We hypothesized that the misfit at TKCH05results from the heterogeneity of the subsurface. To test this hypothesis, we measured four S-wave velocity profiles in the vicinity (< 300 m) of each site with the spectral analysis of surface waves (SASW) method. This KiK-net site pair is ideal for assessing the relative importance of 3D site effects and nonlinear site effects. The linear ground motions at TKCH05 isolate the 3D site effects, as we hypothesized from the linear ground motions and confirmed with our subsequent SASW surveys. The Tokachi-Oki time history at TKCH08 isolates the effects of nonlinearity from spatial heterogeneity because the 3D effects are negligible. The Tokachi-Oki time history at TKCH05 includes both nonlinear and 3D site effects. Comparisons of the accuracy of the SH1D model predictions of these surface time histories from the downhole time histories indicates that the 3D site effects are at least as important as nonlinear effects in this case. The errors associated with the assumption of a 1D medium and 1D wave propagation will be carried into a nonlinear analysis that relies on these same assumptions. Thus, the presence of 3D effects should be ruled out prior to a 1D nonlinear analysis. The SH1D residuals show that 3D effects can be mistaken for nonlinear effects.