We present a simple method to determine gross temporal and spatial characteristics of faulting that can uniquely define the fault plane for a relatively large earthquake. The technique involves deconvolution of teleseismic and regional broadband surface waves (and/or body waves) of a small event (an empirical Green function) from the corresponding signals at the same stations for a large mainshock. This deconvolution corrects each mainshock seismic record for propagation and instrument response, and if the focal mechanisms and centroid depths are identical, deconvolution provides relative source time functions for the mainshock. Azimuthal variations of source time function width or subevent interference reveal the mainshock rupture directivity. Changes in the source mechanism during rupture can be evaluated for some classes of events. Our empirical Green function procedure is novel in its application to surface waves and its use for large (M ≧ 7) earthquakes. We perform synthetic tests to establish intrinsic limitations of the method. The procedure requires very little data processing and can be applied in near-real time with the current distribution of seismic stations that have dial-up data retrieval capability. It also provides a straightforward procedure for reducing bias in source duration estimates resulting from aspherical heterogeneity and for detecting anomalous long-period radiation. We apply the procedure to the 28 June 1992 Landers, California (Mw = 7.3), earthquake. The Landers earthquake source duration lasted about 25 sec, and was dominated by two subevents with predominantly north-northwestward rupture extending 60 to 70 km. The subevents show counter-clockwise rotations of at least 12° in strike that correlate well with mapped surface rupture.