In this study we analyse records from the 'Les Saintes' seismic sequence following the Mw= 6.3 earthquake of 2004 November 11, which occurred close to Guadeloupe (French West Indies). 485 earthquakes with magnitudes from 2 to 6, recorded at distances between 5 and 150 km are used. S-waves Fourier spectra are analysed to simultaneously determine source, path and site terms. The results show that the duration magnitude routinely estimated for the events that occurred in the region underestimate moment magnitude by 0.5 magnitude units over the whole magnitude range. From the inverted seismic moments and corner frequencies, we compute Brune's stress drops. We show that stress drops increase with increasing magnitude. The same pattern is observed on apparent stresses (i.e. the seismic energy-to-moment ratio). However, the rate of increase diminishes at high magnitudes, which is consistent with a constant stress drop model for large events. Using the results of the inversions, we perform ground motion simulations for the entire data set using the SMSIM stochastic simulation tool. The results show that a good fit (σ= 0.25) with observed data is achieved when the source is properly described by its moment magnitude and stress drop, and when site effects are taken into account. Although the magnitude-dependent stress drop model is giving better results than the constant stress drop model, the interevent variability remains high, which could suggest that stress drop depends on other parameters such as the depth of the hypocentre. In any case, the overall variability is of the same order of magnitude as usually observed in empirical ground motion prediction equations.