This study aims at identifying ground‐motion prediction equations (GMPEs) derived from large overseas datasets that may be used for the southwestern part of China (SWC) with or without modifications, identifying the source of misfits and shedding some light on the differences of physical parameters among the three regions, that is, California, Japan, and SWC region. To achieve these objectives, correction functions of physical parameters were added to six GMPEs, including two Next Generation Attenuation‐West (NGA‐West) GMPEs, two NGA‐West2 GMPEs, a GMPE from China, and the GMPE for the shallow crustal and upper‐mantle earthquakes in Japan. The correction coefficients were determined by a reasonably large dataset, including the records from the 2008 7.9 Wenchuan earthquake. However, the data distribution with respect to magnitude is relatively poor and may not be suitable to develop a GMPE based on this dataset only. If the overall correction coefficients for a GMPE are the smallest, this GMPE will be considered as the best GMPE for the SWC data. We found that not all GMPEs can be modified to achieve the same goodness‐of‐fit level; the magnitude scaling rates from most GMPEs differ significantly from those of this dataset, especially at moderately long and long spectral periods; and the aftershock effect does not make a significant and consistent difference. We also found that three GMPEs do not model the site effect well either; the attenuation rates for the SWC region may be similar to those in California and for the shallow crust in Japan; and the magnitude scaling rate and the site class effect in the SWC region are similar to those in Japan. If an overseas GMPE is used for the SWC region, the GMPEs by Zhao, Zhou, et al. (2016) without the normal‐fault term and Abrahamson and Silva (2008) with or without modifications are recommended.