Spatial modeling of ground motion intensity measures (IMs) is required for risk assessment of spatially distributed engineering systems. For example, when a lifeline system is of concern, classical site-specific hazard tools, which treat IMs at different locations independently, may not be adequate to accurately assess the seismic risk. In fact, in this case, modeling of ground motion as a random field is required; it basically consists of assigning a correlation structure to the IM of interest. This work focuses on semiempirical estimation of the correlation coefficient, as a function of intersite separation distance, between residuals with respect to ground motion prediction equations (GMPEs) of horizontal peak ground acceleration (PGA) and peak ground velocity (PGV). In particular, subsets of the European Strong-Motion Database (ESD) and the Italian Accelerometric Archive (ITACA) were employed to evaluate the intraevent residual correlation based on multiple earthquakes, considering different GMPEs fitted to the same records. The analyses were carried out through geostatistical tools, which enabled results to be found that are generally consistent between the two datasets. Correlation for PGV appears to attenuate more gradually with respect to PGA. In order to better understand the dependency of the results on the adopted estimation approach and dataset, some aspects related to the working hypotheses are critically discussed. Finally, estimated correlation models are used to develop illustrative applications of regional probabilistic seismic-hazard analysis.