A detailed error analysis is performed for the calculation of open-ocean salinity changes from carbonate oxygen-18 proxy data. Measurement errors are generally negligible, however, errors due to the goodness of fit of various regressions used in the calculation are significant. Depending on the method used and on the proxies available, error bars (1&sgr;) for the salinity range from 0.6 to 1.8. In particular, expected errors for the midlatitude and tropical salinity differences at the Last Glacial Maximum are around 1.1 to 2.5. The largest errors occur because of the differences between temporal and spatial seawater oxygen-18:salinity relationships and through the use of nonconcurrent temperature measurements. Methods of improving paleosalinity estimates should focus on eliminating these large sources of uncertainty. ¿ 1999 American Geophysical Union