We analyze co-seismic changes of water level in nine wells near Parkfield, California, produced by an M_D 4.7 earthquake on 20 December 1994 in order to test the hypothesis that co-seismic water-level changes are proportional to co-seismic volumetric strain. For each well, a quantitative relationship between water level and volumetric strain can be inferred from water-level fluctuations due to earth tides and barometric pressure. The observed co-seismic water-level changes, which ranged from −16 to +34 cm, can therefore be compared with volumetric strain recorded by borehole strainmeters or calculated using a dislocation model of the earthquake. We were able to find a dislocation model of the earthquake rupture that predicts volumetric expansion at five of the six wells where water level fell co-seismically, as well as volumetric contraction at one of the two sites where water level rose. Strain predicted by the dislocation model is in good quantitative agreement with the strain inferred from water-level changes observed at four of the well sites, as well as strain recorded by three borehole strainmeters. Water-level changes at two more well sites correspond to strain somewhat greater than predicted by the model but agree in sign with model-calculated strains. At three of the well sites, however, water-level changes took place that cannot be explained as responses to co-seismic volumetric strain for any plausible dislocation model of the earthquake rupture. At two of these sites, one in and one near the San Andreas fault, large water-level drops are probably influenced by co-seismic fault creep. The third site has a history of large water-level rises in response to earthquakes at distances up to several hundred kilometers. This data set shows that co-seismic water-level changes in many wells are proportional to volumetric strain but that other wells exist in which different mechanisms dominate co-seismic response.