Short-term earthquake prediction requires sensitive instruments for measuring the small anomalous changes in stress and strain that precede earthquakes. Instruments installed at or near the surface have proven too noisy for measuring anomalies of the size expected to occur, and it is now recognized that even to have the possibility of a reliable earthquake-prediction system will require instruments installed in drill holes at depths sufficient to reduce the background noise to a level below that of the expected premonitory signals. We are conducting experiments to determine the maximum signal-to-noise improvement that can be obtained in drill holes. In a 592 m well in the Mojave Desert near Hi Vista, California, we measured water-level changes with amplitudes greater than 10 cm, induced by earth tides. By removing the effects of barometric pressure and the stress related to earth tides, we have achieved a sensitivity to volumetric strain rates of 10-9 to 10-10 per day. Further improvement may be possible, and it appears that a successful earthquake-prediction capability may be achieved with an array of instruments installed in drill holes at depths of about 1 km, assuming that the premonitory strain signals are, in fact, present. ?? 1985 Birkha??user Verlag.