Controversy over the oxygen isotope composition of seawater began in the 1950's, since which time there has been no agreement over whether the oxygen isotope composition of the oceans has changed over time. Resolving this uncertainty would allow the δ18O values of demonstrably well preserved marine authigenic precipitates to be used to reconstruct surface climate trends back to early Archean times and would help towards a more quantitative description of Earth's global water cycle on geological time scales. Isotopic studies of marine carbonate and silica reveal a trend of increasing δ18O values with decreasing age since the Archean. This trend has been interpreted by some to reflect a progressive increase in seawater δ18O through time; however, it is generally accepted on the basis of ophiolite studies and theoretical considerations that seawater δ18O cannot change significantly because of the buffering effects of ocean crust alteration at mid-ocean ridges. As a result many alternative interpretations have been proposed,including meteoric alteration; warmer paleoclimates; higher seawater pH; salinity stratification and biased sampling. Here we review these interpretations in the light of an updated compilation of marine carbonate δ18O from around the world, covering the Phanerozoic and Precambrian rock records. Recent models of the geological water cycle demonstrate how long-term trends in chemical weathering and hydrothermal circulation can indeed influence the O-isotope composition of the global ocean to the extent necessary to explain the carbonate δ18O trend, with residual variation attributed to climatic fluctuations and post-depositional alteration. We present the further development of an existing model of the geological water cycle. In the model, seawater δ18O increased from about −13.3‰ to −0.3‰ over a period of 3.4 Ga, with average surface temperatures fluctuating between 10 °C to 33 °C, which is consistent with known biological constraints. Similar temperature variations are also obtained, although with higher starting seawater δ18O composition, when more conservative approaches are used that take into account the systematic effects of diagenetic alteration on mean calcite δ18O values. In contrast to much published opinion, the average δ18O value of ocean crust in the model remained relatively unchanged throughout all model runs. Invariable ophiolite δ18O values can, therefore, not be used as a definitive argument against changing seawater δ18O. The most likely explanation for the long-term trend in seawater δ18O invokes two stepwise increases in the ratio of high- to low temperature fluid/rock interactions. An initial increase may have occurred close to the Archean–Proterozoic boundary, but a possibly more significant increase took place near the Proterozoic–Phanerozoic boundary. Possible explanations for extremely low seawater δ18O during the Archean include higher continental weathering rates caused by a combination of higher atmospheric pCO2 (necessarily combined with high CO2 outgassing rates), a greater abundance of relatively easily weathered volcanic rocks in greenstone belts and partial emergence of spreading ridges. The second increase may have been caused by the suppression of low temperature overprinting of ocean crust alteration by the formation of a thick sediment cover on ridge flanks due to the emergence of shelly plankton at the beginning of the Phanerozoic. Postulated increases in spreading ridge depths since the Archean would also have enhanced the efficiency of vertical heat flux and changed the depth at which hydrothermal fluids boil, both of which would favour high- over low-temperature interactions with time.