High resolution strain and tilt recordings were made in the near-field of, and prior to, the May 1983 Coalinga earthquake (ML = 6.7, ?? = 51 km), the August 4, 1985, Kettleman Hills earthquake (ML = 5.5, ?? = 34 km), the April 1984 Morgan Hill earthquake (ML = 6.1, ?? = 55 km), the November 1984 Round Valley earthquake (ML = 5.8, ?? = 54 km), the January 14, 1978, Izu, Japan earthquake (ML = 7.0, ?? = 28 km), and several other smaller magnitude earthquakes. These recordings were made with near-surface instruments (resolution 10-8), with borehole dilatometers (resolution 10-10) and a 3-component borehole strainmeter (resolution 10-9). While observed coseismic offsets are generally in good agreement with expectations from elastic dislocation theory, and while post-seismic deformation continued, in some cases, with a moment comparable to that of the main shock, preseismic strain or tilt perturbations from hours to seconds (or less) before the main shock are not apparent above the present resolution. Precursory slip for these events, if any occurred, must have had a moment less than a few percent of that of the main event. To the extent that these records reflect general fault behavior, the strong constraint on the size and amount of slip triggering major rupture makes prediction of the onset times and final magnitudes of the rupture zones a difficult task unless the instruments are fortuitously installed near the rupture initiation point. These data are best explained by an inhomogeneous failure model for which various areas of the fault plane have either different stress-slip constitutive laws or spatially varying constitutive parameters. Other work on seismic waveform analysis and synthetic waveforms indicates that the rupturing process is inhomogeneous and controlled by points of higher strength. These models indicate that rupture initiation occurs at smaller regions of higher strength which, when broken, allow runaway catastrophic failure. ?? 1987.