VS30, the shear‐wave travel time averaged soil shear‐wave velocity of the top 30 m, has been used to represent site effects in many recent ground‐motion prediction equations (GMPEs). However, while VS30 has been found to be a reasonable parameter to represent site effects in some studies, other studies provide contradictory evidence. In the present study, a systematic comparison between the predictive capabilities of these two site‐effect parameters is carried out using a large ground‐motion dataset from Japan. The basis of the adopted approach is to compare the standard deviations and amplitudes of amplification ratios in empirically modeling site effects by using either site period (TS, four times the shear‐wave travel time from the bedrock to the ground surface) or VS30. The site effects modeled specifically include site amplification ratios between surface and borehole records from KiK‐net, in addition to the site‐effect terms from a GMPE. For KiK‐net data, TS is determined to be a better predictive parameter than VS30 for soil sites with TS>0.6 s, while the two parameters lead to a similar variability in amplification ratios for sites with TS<0.6 s. For site effects obtained from the GMPE, VS30 and TS are statistically equal for all site classes at most periods, while VS30 leads to smaller variability than TS at some spectral periods. The conflict between the KiK‐net surface–borehole records, and the results from the GMPE is likely to be a result of large variability in the GMPE, containing source‐, path‐, and site‐variability, as compared with the reduced variability in the surface–borehole KiK‐net data pairs. Although VS30 and TS lead to statistically similar standard deviations for the data from a GMPE, TS still leads to better median amplification ratios than VS30.