Renewed earthquake activity at Blue Mountain Lake (BML), New York, in July 1973 provided an excellent opportunity to monitor the travel time ratio of S to P waves (ts/tp) in real time and to test the ts/tp technique as a predictive tool. From a mean value of 1.73 on July 30, 1973, ts/tp decreased to about 1.5 over the next 2-3 days. On August 1 a prediction was made that an earthquake of magnitude 2.5-3 would occur in a few days. Upper limits of the magnitude and the time of occurrence of the expected earthquake were inferred from the spatial extent to the seismic anomaly. As a result of the prediction an additional strong motion accelerograph (SMA) was installed in the source region. At 2310 UT on August 3, 1973, a magnitude 2.6 earthquake occurred at BML and triggered two SMA's. In addition to the seismic activity a number of explosions were recorded from a variety of azimuths. The P wave arrivals from distant quarry blasts, refracted from a high-velocity layer at 4 km beneath BML, showed late arrivals at five stations during the premonitory low in ts/tp. The P wave delays were maximum (0.13 s) in the hypocentral region of the earthquake and decreased away from it along two profiles. These results indicate that changes in ts/tp are caused by changes in the material properties of the earthquake source region. The anomalous zone (region of low P velocity) for the August 3 earthquake (aftershock length 1 km, depth 1¿0.1 km) was about 3-5 km in radius and was wholly or largely limited to the layer above the interface at 4-km depth. In contrast to the P delays observed from distant quarry blasts. P and S arrivals from local construction blasts (Δ<10 km) show no large premonitory changes in either P or S travel times. This observation suggests that either no significant velocity anomalies occurred in the upper 0.5 km of the BML region or that the dilatant cracks were predominantly horizontal, the result being strong velocity anisotropy. This conclusion is supported by data from seismic sources; events located at shallow depths show normal (1.75) to high (1.85) ts/tp values even when values as low as 1.5 are observed from deeper sources. This finding places constraints on the use of artificial sources to monitor changes in velocity, since the effects of anisotropy may have to be taken into account in addition to ensuring that ray paths to the recording stations penetrate the zone of anomalous velocity. A maximum likelihood method was used to invert data from individual small earthquakes to determine P and S velocities in the anomalous zone. The results, which are consistent with the explosion data, indicate that the premonitory decreases in P and S velocities were much more pronouced for sources at depths of 1-2 km than for those near the surface. The inferred low values of P and S velocities were, respectively, about 22 and 12% below normal. This study also shows that ts/tp inferred from a Wadati plot is a function of the distribution of stations relative to the anomalous zone. To optimize the use of ts/tp as a predictive tool, this study suggests that not only temporal but spatial variations of velocity anomalies should be monitored. This emphasizes the need for multistation coverage for reliable earthquake prediction.