This study evaluated the source, path, and site effects of the vertical ground motions from the western and the southwestern parts of China (referred to as SWC hereafter) using 2403 records from 449 earthquakes, including the records from the 2008 7.9 Wenchuan earthquake and its aftershocks. Only 677 records are from 73 mainshocks, and 259 events do not have a known focal mechanism. There is a large magnitude gap in the dataset, for example, there is only one event between 6.3 and 7.8. The average numbers of records per recording station and per earthquake are small, and many sites do not have a measured shear‐wave velocity profile. These shortcomings make it difficult to develop a robust ground‐motion prediction equation (GMPE) without adding overseas data or using a reference GMPE developed from a large dataset. We compared the SWC dataset with five recent GMPEs, three based on the Next Generation Attenuation‐West2 dataset, one based on Europe and the Middle East, and one based on the shallow‐crustal and upper‐mantle earthquakes in Japan. We decomposed the total residuals for each model into constant term, between‐event, and within‐event residuals and calculated the corresponding standard deviations. The maximum log likelihood and the standard deviations suggest that, among the five GMPEs, the Zhao et al. (2017) model without the normal‐fault term may be the most suitable GMPE for a probabilistic seismic hazard study in the SWC region. Correction functions based on simple magnitude, path, and site effect parameters were used to correct the residuals and to obtain the leftover between‐ and within‐event standard deviations. These standard deviations appear to suggest that the GMPE from Zhao et al. (2017) without a normal‐fault term may be the most suitable reference GMPE for developing a new GMPE for the SWC region.