Site effects for hard‐rock sites are typically computed using analytical models for the effect of κ0, the high‐frequency attenuation parameter. New datasets that are richer in hard‐rock recordings allow us to evaluate the scaling for hard‐rock sites (e.g., VS30>1500 m/s). The high‐frequency response spectra residuals are weakly correlated with κ0, in contrast to the strong scaling with κ0 in the analytical models. This may be due to site‐specific shallow resonance patterns masking part of the effect of attenuation due to damping. An empirical model is developed for the combined VS30 and κ0 scaling for hard‐rock sites relative to a reference site condition of 760 m/s (i.e., correction factors that should be used for going from soft rock to hard rock, taking into account the net effect of VS and κ0). This empirical model shows high‐frequency amplification that is more similar to the analytical prediction corresponding to a hard‐rock κ0 of 0.020 s rather than the typical value of 0.006 s, which is commonly used for hard‐rock sites in the central–eastern United States. Compared to the current analytical approach, this leads to a reduction of high‐frequency (>20 Hz) scaling of about a factor of 2.