This study investigates the theoretical improvement of the ground-motion models used for ShakeMap implementation in Italy, obtainable by accounting for regional differences of wave propagation and for site effects. The analysis considers 922 recordings of 116 earthquakes with local magnitude ML in the range 2.7–4.5, collected from 137 stations at epicentral distances up to 100 km. Alternative partitions of the available stations divided into separated zones are considered. For each zone in a partition, an empirical ground-motion model is estimated, which includes station correction terms. A genetic algorithm is used in order to minimize the standard deviation of the residuals log10(observed)-log10(predicted) over the set of the possible partitions. For the logarithm of the peak ground acceleration, it is estimated that site effects contribute to about 30% of the overall standard deviation, while regional differences contribute with no more than 4%. Furthermore, the estimated station correction terms are weakly correlated with the amplification factors based on shear-wave velocity in the upper 30 m (VS30) used in ShakeMap to model site effects. Similar results apply to peak ground velocity and spectral acceleration response at 0.3, 1, and 3 sec. These results suggest that parametrizations of site effects alternative to those based on VS30 should be considered, while the contribution of any regionalization of the ground-motion models is marginal.