Local earthquakes with magnitudes of >2.5 and within 20--50 km of tiltmeters along the San Andreas fault typically generate offsets in tilt, tilt seismograms, and impulsive tilt behavior at the time of the earthquake. The amplitudes and azimuths of the coseismic tilt offsets from local earthquakes observed at a small array of four instruments approximately 30 km south of Hollister, California, are compared to the amplitudes and azimuths predicted by a rectangular source, elastic half-space dislocation model. Almost all observed coseismic offset amplitudes are 1--3 orders of magnitude larger than the predicted amplitudes. The predicted offset azimuths also are not in agreement with the observed azimuths. There is neither a uniform method of scaling the predicted amplitudes nor a constant rotation that may be applied to the predicted azimuths that will consistently reproduce the observed offsets. Errors in hypocenter location and fault plane orientation are not sufficient to explain the discrepancies between observations and predictions. Similar results occur for teleseismic earthquakes. A lack of agreement in the observed offset amplitudes across the array indicates that tilt changes are triggered at or near the instrument site by the passage of seismic waves. No significant agreement was obtained between the direction of coseismic tilts and either the secular tilt trends or local geologic features. Triggered movement on near-surface cracks, fractures, and minor faults appears the most viable physical explanation for the observed offsets. Short-base-line near-surface tiltmeters appear inadequate for measuring tilt displacement fields generated by local earthquakes. Deep borchole installations appear necessary for this measurement. Coseismic tilt transients appear to be caused by seismically induced water table perturbations near the tiltmeter site.