We have conducted a series of cross-well experiments to continuously measure in situ temporal variations in seismic velocity at two test sites: building 64 (B64) and Richmond Field Station (rfs) of the Lawrence Berkeley National Laboratory in California. A piezoelectric source was used to generate highly repeatable signals, and a string of 24 hydrophones was used to record the signals. The B64 experiment was conducted utilizing two boreholes 17 m deep and 3 m apart for ∼160 h. At rfs, we collected a 36-day continuous record in a cross-borehole facility using two 70-m-deep holes separated by 30 m. With signal enhancement techniques we were able to achieve a precision of ∼6.0 nsec and ∼10 nsec in delay-time estimation from stacking of 1-hr records during the ∼7- and ∼35-day observation periods at the B64 and rfs sites, which correspond to 3 and 0.5 ppm of their travel times, respectively. Delay time measured at B64 has a variation of ∼2 μsec in the 160-hr period and shows a strong and positive correlation with the barometric pressure change at the site. At rfs, after removal of a linear trend, we find a delay-time variation of ∼2.5 μsec, which exhibits a significant negative correlation with barometric pressure. We attribute the observed correlations to stress sensitivity of seismic velocity known from laboratory studies. The positive and negative sign observed in the correlation is likely related to the expected near- and far-field effects of this stress dependence in a poroelastic medium. The stress sensitivity is estimated to be ∼10−6/ Pa and ∼10−7/Pa at the B64 and rfs site, respectively.