Daily averages of magnetic difference fields for the years 1974--1980 between 23 magnetometers on the San Andreas fault (between San Francisco and the Mexican border) have been analyzed to isolate the geomagnetic variation of crustal origin and to test if it is related to stress. Field changes due to nonuniform secular variation from currents in the core have been identified and removed. External currents in the ionosphere and magnetosphere induced changes in the difference fields due to contrasting impedances at magnetometer sites. These changes are identified and removed by multichannel predictive filtering using Boulder magnetic observatory component fields as inputs to the prediction operator. The remaining fluctuations (up to several nT) are attributed to crustal magnetization changes. Evidence that these changes are stress related includes the variation prior to the Busch fault earthquake (1974) seen at the two stations closest to (within 10 km) the epicenter. Also, stations within 50 km of the 1979 Coyote earthquake undergo localized field change up to a year before that event. If truly precursive, these period 1975--1978 the geomagnetic variation is approximately uniform across the array, without the development of any significant anomalies. This period roughly coincides with the 1975.7--1978.7 seismic lull in nearly all of California. It is followed by a marked increase in crustal secular variation. Long-range coherence between anomalous variation at different stations is not seen, other than the variation before the Busch fault earthquake, suggesting that localized variation occurs on a scale of less than 10 km. Coseismic tectonomagnetic events are not observed at the times of the Busch and Coyote earthquakes, but model calculations suggest that observable events are not expected. If the observed changes are due to stress, they indicate that changes of aseismic stress are significantly larger than coseismic stress release.