The standard deviation of ground‐motion prediction equation under the ergodic assumption is commonly used for the site‐specific probabilistic seismic hazard analysis. However, this assumption leads to overestimations of site‐specific ground‐motion variability. The single‐station standard deviation () after removing the ergodic assumption on site response is therefore proposed for evaluating site‐specific ground‐motion variability. In this study, we evaluate for the seismically active Sichuan region, China. Based on strong‐motion data in this region, we first establish prediction equations for the sole purpose of computing means of ground‐motion intensity measures (peak ground acceleration [PGA] and pseudospectral accelerations [PSAs]) for subsequent analyses. Then, we apply 1463 recordings at a total of 47 stations to compute the values. The standard deviations under the ergodic assumption range from 0.67 to 0.86 on the natural logarithmic scale, whereas values range from 0.59 to 0.72, indicating the reduction after removing the ergodic assumption on site response. The event‐corrected single‐station standard deviations () are reduced by about 15%–25% compared with the within‐event standard deviation. The values calculated in this study are generally higher than the usual values reported by previous studies. More heterogeneous propagation medium in this region, the boundary of the Sichuan basin and the eastern margin of the Tibet plateau, may cause larger values than usual results. We infer that the use of aftershock recordings may be another reason for the large values in that ruptures of mainshocks may amplify upper crust heterogeneity where seismic waves propagate during aftershocks. Therefore, the larger values in this study could be a sort of warning against the usually assumed regional independency of based on the usual values from previous studies.