The validity and the stability of a ground-motion simulation method based on the recordings of a single small event as an empirical Green’s function (EGF) is tested on a seismic crisis that occurred 25 km offshore of the Guadeloupe Islands (Caribbean arc). We aim to determine if (1) the method enables us to reproduce the observed ground motion, (2) the choice of the small event taken as an EGF is crucial for the simulations, and (3) the method provides valuable results compared with ground-motion prediction equations (GMPEs). We have successively used the recordings of 10 small earthquakes (Mw 4.2–5.1) to simulate the ground motions generated by the mainshock (Mw 6.4), at 12 accelerometric stations. We first determined the moment and focal mechanisms of the 10 events chosen as an EGF, as well as the stress-drop ratio C between each of these events and the mainshock. Then, we simulated 500 accelerograms for each EGF and each station. A good reproduction of the mainshock response spectra, the peak ground acceleration, and the duration of the signal was obtained using 9 out of 10 EGFs. For stations with site effects, the results obtained are much closer to the real data than values given by the GMPEs on sediment sites. In the case of blind predictive simulation, we propose to calibrate the stress-drop ratio C through a comparison between the simulated response spectra on rock site stations and the values predicted by GMPEs.