It has become a widespread practice to convert δ18Op values measured in human and animal dental enamel to a corresponding value of δ18Ow and compare these data with mapped δ18Ow groundwater or meteoric water values to locate the region where the owner of the tooth lived during the formation of the enamel. Because this is a regression procedure, the errors associated with the predicted δ18Ow values will depend critically on the correlation between the comparative data used to perform the regression. By comparing four widely used regression equations we demonstrate that the smallest 95% error is likely to be greater than ±1% in δ18Ow, and could be as large as ±3.5%. These values are significantly higher than those quoted in some of the recent literature, and measurements with errors at the higher end of this range would render many of the published geographical attributions statistically unsupportable. We suggest that the simplest solution to this situation is to make geographical attributions based on the direct comparison of measured values of δ18Op rather than on predicted values of δ18Ow. Am J Phys Anthropol, 2011. © 2011 Wiley-Liss, Inc.