The Landers earthquake (M_W 7.3) of 28 June 1992 had a very emergent onset. The first large amplitude arrivals are delayed by about 3 sec with respect to the origin time, and are preceded by smaller-scale slip. Other large earthquakes have been observed to have similar emergent onsets, but the Landers event is one of the first to be well recorded on nearby stations. We used these recordings to investigate the spatial relationship between the hypocenter and the onset of the large energy release, and to determine the slip function of the 3-sec nucleation process. Relative location of the onset of the large energy release with respect to the initial hypocenter indicates its source was between 1 and 4 km north of the hypocenter and delayed by approximately 2.5 sec. Three-station array analysis of the P wave shows that the large amplitude onset arrives with a faster apparent velocity compared to the first arrivals, indicating that the large amplitude source was several kilometers deeper than the initial onset. An M_L 2.8 foreshock, located close to the hypocenter, was used as an empirical Green's function to correct for path and site effects from the first 3 sec of the mainshock seismogram. The resultant deconvolution produced a slip function that showed two subevents preceding the main energy release, an M_W4.4 followed by an M_W 5.6. These subevents do not appear anomalous in comparison to simple moderate-sized earthquakes, suggesting that they were normal events which just triggered or grew into a much larger earthquake. If small and moderate-sized earthquakes commonly “detonate” much larger events, this implies that the dynamic stresses during earthquake rupture are at least as important as long-term static stresses in causing earthquakes, and the prospects of reliable earthquake prediction from premonitory phenomena are not improved.