The 1990 Upland earthquake was one of the first sizable local events to be recorded broadband at Pasadena, where the Green's functions appropriate for the path are known from a previous study. The synthetics developed in modeling the 1988 Upland sequence were available for use in rapid assessment of the activity. First-motion studies from the Caltech-USGS array data gave two solutions for the 1990 main shock based on the choice of regional velocity models. Although these focal mechanisms differ by less than 5° in strike and 20° rake, it proved possible to further constrain the solution using these derived Green's functions and a three-component waveform inversion scheme. We obtain from long-period waves a fault-plane solution of θ = 216°, δ = 77°, λ = 5.0°, M0 = 2.5 × 1024 dyne-cm, depth = 6 km, and a source duration of 1.2 sec, for which the orientation and source depth are in good agreement with the first-motion results of Hauksson and Jones (1991). Comparisons of the broadband displacement records with the high-pass Wood-Anderson simulations suggests the 1990 earthquake was a complicated event with a strong asperity at depth. Double point-source models indicate that about 30 per cent of the moment was released from a 9-km deep asperity following the initial source by 0.0 to 0.5 sec. Our best-fitting distributed fault model indicates that the timing of our point-source results is feasible assuming a reasonable rupture velocity. The rupture initiated at a depth of about 6 km and propagated downward on a 3.5 by 3.5 km (length by width) fault. Both the inversion of long-period waves and the distributed fault modeling indicate that the main shock did not rupture the entire depth extent of the fault defined by the aftershock zone. A relatively small asperity (about 1.0 km2) with a greater than 1 kbar stress drop controls the short-period Wood-Anderson waveforms. This asperity appears to be located in a region where seismicity shows a bend in the fault plane.