In the absence of onsite recordings, the principal approach for site‐specific analysis of site response uses the assumption of shear waves propagating vertically through horizontal soil layers (i.e., ground response analysis [GRA]). We consider the fundamental issue of the effectiveness of GRA methods using a California dataset of 21 vertical arrays. Transfer functions derived from relatively weak‐motion data (for which soil nonlinear effects are expected to be small) are compared to predictions from linear ground response analyses performed using three damping models—geotechnical models derived from laboratory element tests, models for quality factor () based on seismological inversion, and models derived from the site‐specific site decay parameter (). When compared to prior results for KiK‐net sites in Japan by others, the California sites have, on average, improved match of empirical and theoretical transfer function shapes and more event‐to‐event consistency. When applied in aggregate, using ‐informed damping results in a slightly better fit between predicted and observed transfer functions than alternative damping models. As observed previously, the geotechnical models underestimate site attenuation, whereas the models overestimate site attenuation.