Abstract: Different parameters are used for stipulating site classes in seismic design codes of different countries of the world, among which the schemes from China, Japan and U.S.A are the most representative. Few studies have been examined and evaluated whether the site classification scheme can functionally reflect the intra-class uniformity and inter-class separateness of ground motion site response. To examine the performance of site classification schemes, the strong motion records and borehole data from KiK-net seismograph networks are used, the sites of strong motion station are classified respectively according to the schemes from the three countries, and the site amplification factors are calculated based on the method of residual analysis of ground motion model. The mean and standard deviation of site amplification factors for each site class are compared among different classification schemes. The results show that: For PGA and short period (T < 0.1 s) ground motion, the scheme from seismic code of Japan which uses site nature period (T_S), does not reflect clearly the inter-class separateness; while the schemes of China and U.S.A. have better performance, because soil covering depth and equivalent shear wave velocity (V_SE) are used in China and V_S30 is used in U.S.A. In general, the scheme from seismic code of Japan is the best one in reflecting intra-class uniformity; but for long period (T > 2 s) ground motion, the one from China performs best. Great variability of site amplification of medium-long period ground motions is observed for site class III defined by seismic code of China, calling for a further investigation on the rationality of its site classification scheme. The results achieved can provide theoretical support for improving site classification scheme in future.