A numerical model which describes oxygen isotope exchange during burial and recrystallization of deep-sea carbonate is used to obtain information on how sea surface temperatures have varied in the past by correcting measured δ18O values of bulk carbonate for diagenetic overprinting. Comparison of bulk carbonate and planktonic foraminiferal δ18O records from ODP site 677A indicates that the oxygen isotopic composition of bulk carbonate does reflect changes in sea surface temperature and δ18O. At ODP Site 690, we calculate that diagenetic effects are small, and that both bulk carbonate and planktonic foraminiferal δ18O records accurately reflect Paleogene warming of high latitude surface oceans, biased from diagenesis by no more than 1°C. The same is likely to be true for other high latitude sites where sedimentation rates are low. At DSDP sites 516 and 525, the effects of diagenesis are more significant. Measured δ18O values of Eocene bulk carbonates are more than 2% lower at deeply buried site 516 than at site 525, consistent with the model prediction that the effects of diagenesis should be proportional to sedimentation rate. Model-corrections reconcile the differences in the data between the two sites; the resulting paleotemperature reconstruction indicates a 4°C cooling of mid-latitude surface oceans since the Eocene. At low latitudes, the contrast in temperature between the ocean surface and bottom makes the carbonate δ180 values particularly sensitive to diagenetic effects; most of the observed variations in measured δ18O values are accounted for by diagenetic effects rather than by sea surface temperature variations. We show that the data are consistent with constant equatorial sea surface temperatures through most of the Cenozoic, with the possible exception of the early Eocene, when slightly higher temperatures are indicated. We suggest that the lower equatorial sea surface temperatures for the Eocene and Oligocene reported in other oxygen isotope studies are artifacts of diagenetic recrystallization, and that it is impossible to reconstruct accurately equatorial sea surface temperatures without explicitly accounting for diagenetic overprinting.