The full characterization of a seismic source requires the specification of the hypocentre location and the source mechanism. The non-linear inversion is accomplished in two stages using the Neighborhood Algorithm (NA), a direct search method in parameter space which is able to preferentially sample those regions with least data misfit; there are just two control parameters and no differentiation is employed. The first step is hypocentre location using a 4-D search space and a focussed search strategy. The second step is waveform inversion of the early part of P and S wavetrains to refine source depth and extract the source mechanism. With a moment tensor representation, this second stage explores an 8-D parameter space. For both the hypocentre and source mechanism inversion the NA method provides rapid and effective results with information from just a few stations, as illustrated with an event in southern Xinjiang. ¿ 2000 American Geophysical Union