As part of the Parkfield, California, earthquake prediction experiment, water level in monitored in a well 460 m from the main trace of the San Andreas fault on Middle Mountain, in the preparation zone of the anticipated Parkfield earthquake. The well configuration allows water level to be monitored in two fluid reservoirs at depths of 85 and 250 m below land surface. During 1987, water level changes were recorded during 12 of the 18 episodes of accelerated fault creep detected by a creep meter spanning the fault trace 750 m northwest of the well. The creep-related water level changes in the shallow reservoir have durations of less than 1 day, whereas in the deeper reservoir the changes persist for as long as 2 months. These data suggest that the transient nature of the water level changes in the shallow interval is due to vertical flow to the water table and is not evidence that creep events propagate past the well. Phase leads of earth tidal constituents in the water level data from the shallow interval relative to the same constituents in the local volume strain tide support the interpretation of significant flow to the water table at periods of 1 day or less. The form of the water level changes in the deep interval is affected by horizontal flow to the well bore. This effect can be removed from the water level records using a theoretical response curve constrained by the phases of earth tidal constituents in the deep interval relative to the local volume strain tide. For the events where the strain tide. For the events where the signal in the shallow interval has been large enough to measure, the sizes of the simultaneous water level changes in the two reservoirs are consistent with the same amounts of volume strain occurring at both depths. ¿ American Geophysical Union 1989