Quantification of regional seismic risk is based on spatially correlated random fields and requires modeling of the joint distribution of ground‐motion intensity measures at all sites of interest. In particular, when a portfolio of buildings or a transportation/distribution network (lifeline) is of concern, correlation models for elastic spectral acceleration (SA) may also be required in order to estimate the expected loss in case of seismic events. The presented study focuses on semi‐empirical estimation of spatial correlation as a function of intersite separation distance. In fact, this paper complements, and is based on, preceding work of the authors referring to spatial correlation of peak ground acceleration and velocity (Esposito and Iervolino, 2011). The evaluation of correlation for ground‐motion residuals was performed on data from multiple earthquakes, considering different ground‐motion prediction equations fitted to the same records. Correlation analyses, carried out through geostatistical tools, considered two datasets: the Italian Accelerometric Archive and the European Strong‐Motion Database. Results appear generally consistent with previous research on the same topic. Finally, simple relationships providing the correlation range of intraevent residuals of SA, as a function of structural period, were derived for each dataset. The developed models are useful for earthquake engineering applications where spatial correlation of peak ground motion is required.