Geochemical analyses of coral skeletons are increasingly used to estimate past sea surface temperatures (SSTs). In this paper we suggest that the standard method of calibrating geochemical time series against a (usually short) local time series requires modification. In order to draw large-scale inferences about climate from coral proxy data it is also necessary to (1) calibrate against larger fields such as the local gridded data sets and (2) validate results against an independent data set (e.g., early 20th century). This approach has been applied in a pilot study to a coral record from New Caledonia. Despite a high Δ18O correlation (r=-0.88) with the in situ and gridded SST data sets, estimated early 20th-century temperatures are more than 1.5 ¿C colder than observed if the standard seasonal calibration is used. Regression against mean annual temperatures, which has a different slope relation, yields better estimates of early 20th-century SSTs. However, testing of a Sr/Ca record from New Caledonia yields better agreement with early 20th century SSTs. Routine validation exercises for other coral sites are necessary to clarify the robustness of geochemical coral proxies as estimators of past environmental change. ¿ 1999 American Geophysical Union