The general failure of searches for precursory seismic travel-time variations associated with strike-slip earthquakes in California has led to this investigation into the feasibility of using a controlled-source seismic method to improve significantly the precision of travel-time measurements, and to investigate the nature of any detected travel-time changes. Travel times have been measured over a period of several years at sites south of Hollister, California, along the seismically active creeping zone of the San Andreas fault, using a single-channel VIBROSEIS system with real-time on-site data processing. At a site near Bear Valley, 7 km from the fault, no variations in the travel time of a deep crustal reflection were observed that could be associated with local earthquakes. However, significant variations (0.5 to 2.5 msec) of first arrival travel times observed near the Cienega Winery may have been associated with a series of nearby earthquakes and with a creep event. Major sources of measurement error have been identified in source variations and in rainfall-induced variations in near-surface properties. The former limits the precision of the deep reflection measurements to about 0.05 per cent of the travel time and the first arrival measurements to about 0.1 per cent of the travel time. The second effect is apparent as seasonal oscillations in travel time of as much as 15 to 20 msec, and also in wavelet amplitude and waveform, giving an implied travel-time accuracy of about 0.2 per cent for the deep reflection measurements and about 1 per cent for first arrivals. While these noise levels are disappointing, they can be reduced significantly by improved field procedures. Ongoing experiments are testing such procedures.