The Coyote Lake earthquake (ML = 5.7) occurred within the dense network of seismic instruments of central California and was recorded by numerous accelerograph stations. Two of these stations, located within the fault zone itself, provide particularly remarkable records of the rupture process. We use these data obtained in the immediate vicinity of the fault and the broadband seismograms recorded at Berkeley, about 100 km away from the epicenter, to infer the velocity of propagation of the rupture, the extent of the fractured area, the rupture front geometry, and other characteristics of the fracture mechanism. In order to extract this information from the data, we model the earthquake as a propagating dislocation starting at the hypocenter and spreading radially. We compute the resulting ground motion at the receiver sites using the discrete wavenumber representation of Green's functions (Bouchon, 1981). The results yield a rupture velocity of 2.6 km/sec, a fault length of about 14 km and a fault slip of 15 to 20 cm. They also show that the fault extends to shallower depth than indicated by the aftershocks. The most important result is the finding that slip, at any given location on the fault, takes place in a very short time span and that simple uniform dislocation models give a good description of the rupture process.