Monitoring of the water level has been continued in 100-m-deep and 500-m-deep wells in the southern Kanto District, Japan, to examine the relationship between changes in the well water level and earthquake occurrence. By using a Bayesian approach, the observed data were decomposed into four components: tidal response, barometric response, trend (drift), and irregular (noise) components. Loading effects of the ocean tide on aquifer dilatation were evaluated by using Schwiderski's ocean tide models and the harmonic constants compiled by the International Hydrographic Bureau. Well-aquifer properties were examined based on the tidal and barometric responses of the water level. Coseismic water level steps were detected in both wells for two earthquakes during the observation period from April 1985 to December 1989. By calibrating the strain sensitivity of the well water levels based on the tidal responses, strain changes equivalent to the coseismic steps in the water levels were estimated. The coseismic strain changes obtained for the deeper well can be explained reasonably by earthquake fault models in a viscoelastic half-space. However, those observed in the shallower well were too large and in the wrong direction to be explained by the earthquake fault models. ¿1991 American Geophysical Union