Near-source strong motion velocity records and teleseismic short-period P waveforms are modeled to obtain the spatial and temporal distribution of slip for the 1984 Morgan Hill earthquake. Both forward modeling and constrained, least-squares inversion techniques are used to interpret the strong motion velocity waveforms in the frequency range of approximately 0.2 to 2.0 Hz. These data support a nearly unilateral rupture to the southeast with a rupture propagation velocity of nine-tenths of the local S-wave velocity. The majority of the slip occurs over a fault length of 25 km and to a first approximation can be interpreted as two main source regions, each with an extent of about 5 km with their centers separated by about 12 km. However, each of the sources has detailed structure of its own, and a simple two-point-source model is not an accurate representation of the Morgan Hill earthquake. The second source occurs about 4.5 sec after the first and is approximately 3 times larger. The maximum dislocation on the fault plane is about 1 m. The total moment of the earthquake is estimated to be 2.1 ×1025 dyne-cm. The Morgan Hill earthquake offers convincing evidence for very inhomogeneous slip and stress distributions on shallow strike-slip faults.