Spatial correlation of intraevent peak ground-motion amplitudes at different sites is an important issue for seismic hazard and risk assessment of spatially distributed buildings and infrastructure. Correlated seismic effects cause acute concentration and accumulation of seismic losses, potentially resulting in a catastrophic event. We investigate the adequacy of the existing spatial correlation model of Goda and Atkinson (2009b) for Japanese earthquakes in light of strong ground motion data obtained from the SK-net database, which achieves greater spatial density of recording stations in the Kanto region. The SK-net dataset is suitable for filling in gaps in the Goda–Atkinson dataset at short separation distances where data coverage is sparse. Spatial correlations based on the SK-net data decrease gradually with increasing interstation separation distance. At short separation distances, the estimated spatial correlation data points show large variability around the average trend. Sensitivity to the manner of estimating intraevent standard deviation is significant, as positive correlations among residuals occur due to the dense spatial coverage of recording stations. The developed spatial correlation model fits the observed data reasonably well and is consistent with the Goda–Atkinson model. At short separation distances (less than 1 km), wherein empirical data are limited and estimates are uncertain, discretion is required in adopting such models for seismic hazard and risk assessment of spatially distributed structures.