We examine the initial subevent (ISE) of the M?? 6.7, 1994 Northridge, California, earthquake in order to discriminate between two end-member rupture initiation models: the 'preslip' and 'cascade' models. Final earthquake size may be predictable from an ISE's seismic signature in the preslip model but not in the cascade model. In the cascade model ISEs are simply small earthquakes that can be described as purely dynamic ruptures. In this model a large earthquake is triggered by smaller earthquakes; there is no size scaling between triggering and triggered events and a variety of stress transfer mechanisms are possible. Alternatively, in the preslip model, a large earthquake nucleates as an aseismically slipping patch in which the patch dimension grows and scales with the earthquake's ultimate size; the byproduct of this loading process is the ISE. In this model, the duration of the ISE signal scales with the ultimate size of the earthquake, suggesting that nucleation and earthquake size are determined by a more predictable, measurable, and organized process. To distinguish between these two end-member models we use short period seismograms recorded by the Southern California Seismic Network. We address questions regarding the similarity in hypocenter locations and focal mechanisms of the ISE and the mainshock. We also compare the ISE's waveform characteristics to those of small earthquakes and to the beginnings of earthquakes with a range of magnitudes. We find that the focal mechanisms of the ISE and mainshock are indistinguishable, and both events may have nucleated on and ruptured the same fault plane. These results satisfy the requirements for both models and thus do not discriminate between them. However, further tests show the ISE's waveform characteristics are similar to those of typical small earthquakes in the vicinity and more importantly, do not scale with the mainshock magnitude. These results are more consistent with the cascade model.