A key step in probabilistic seismic-hazard assessment is the prediction of expected ground motions produced by the seismic sources. Most probabilistic studies use a ground-motion prediction model to perform this estimation. The present study aims at testing the use of simulations in the probabilistic analysis instead of ground-motion models. The method used is the empirical Green’s function method of Kohrs-Sansorny et al. (2005), which takes into account the characteristics of the source, propagation paths, and site effects. The recording of only one small event is needed for simulating a larger event. The small events considered here consist of aftershocks from the M 6.4 Les Saintes earthquake, which struck the Guadeloupe archipelago (French Antilles) in 2004. The variability of the simulated ground motions is studied in detail at the sites of the French Permanent Accelerometric Array. Intrinsic variability is quantified: ground motions follow lognormal distributions with standard deviations between 0.05 and 0.18 (log units) depending on the spectral frequency. One input parameter bearing large uncertainties is the ratio of the stress drop of the target event to the small event. Therefore, overall sigma values (and medians) are recomputed, varying stress drop ratio values between 1 and 15. Sigma values increase but remain in general lower or equal to the sigma values of current ground-motion prediction models. A simple application of this hybrid deterministic–probabilistic method is carried out at several sites in Guadeloupe for the estimation of the hazard posed by an M 6.4 occurring in the rupture zone of the Les Saintes event.