Measuring stress changes within seismically active fault zones has been a longsought goal of seismology. Here we show that such stress changes are measurable by exploiting the stress dependence of seismic wave speed from an active source cross-well experiment conducted at the SAFOD drill site. Over a two-month period we observed an excellent anti-correlation between changes in the time required for an S wave to travel through the rock along a fixed pathway -a few microseconds--and variations in barometric pressure. We also observed two large excursions in the traveltime data that are coincident with two earthquakes that are among those predicted to produce the largest coseismic stress changes at SAFOD. Interestingly, the two excursions started approximately 10 and 2 hours before the events, respectively, suggesting that they may be related to pre-rupture stress induced changes in crack properties, as observed in early laboratory studies [1][2] .It is well known from laboratory experiments that seismic velocities vary with the level of applied stress [3][4][5] . Such dependence is attributed to the opening/closing of microcracks due to changes in the stress normal to the crack surface [6][7][8] . In principle, this dependence constitutes a stress meter, provided the induced velocity changes can be 2 measured precisely and continuously. Indeed, there were several attempts in the 1970s to accomplish this goal using either explosive or non explosive surface sources 9-11 . The source repeatability and the precision in traveltime measurement appeared to be the main challenges in making conclusive observations.With the availability of highly repeatable sources, modern data acquisition systems, and advanced computational capability, Yamamura et al. 12 showed compelling evidence that seismic velocity along a baseline in a vault near the coast of Miura Bay, Japan, responds regularly to tidal stress changes. Silver et al. 13 found an unambiguous dependence of seismic velocity on barometric pressure from a series of cross-well experiments at two test sites in California. The stress sensitivity depends primarily on crack density and has a strong nonlinear dependence on confining pressure.Consequently, crack density is expected to decrease rapidly with depth as should stress sensitivity. It is thus unclear whether the stress-induced velocity variations observed at shallow depths [12][13] are still detectable at seismogenic depth.To explore stress sensitivity at seismogenic depth, we have conducted an experiment at Parkfield where adjacent deep wells, the SAFOD (San Andreas Fault Observatory at Depth) pilot and main holes (Figure 1), are available. Accurately located seismicity together with the availability of high-quality geophysical data in the Parkfield region make it one of the best areas to detect temporal changes related to the earthquake cycle.A specially-designed 18-element piezoelectric source and a three-component accelerometer were deployed inside the pilot and main holes, respectively, at ~1 km depth (see methods)...