We determine the structure of a network of sites from which coral-based, proxy measurements of sea surface temperature (SST) variability minimize the error in a reconstruction of the large-scale features of the global SST field. For a wide range of coral-derived SST observational error and several minimization criteria, sites in the equatorial oceans, especially the central and eastern equatorial Pacific, best minimize the error in the reconstruction. If the observational error is sufficiently low, additional optimal sites are in selected subtropical locations. If the observational error is high, the error is minimized by resampling the most important equatorial sites. The marginal return on additional sites diminishes rapidly and is not sensitive to the size of the observational error: two sites reduce the analysis error by half as much as 10 sites, while the first 6--7 sites achieve half the error reduction of all 65 sites in the analysis domain. In the extratropics the reduction of reconstruction error is small and gradual, while error in the tropical Pacific is sharply reduced by 2--3 sites and gradually thereafter. These results suggest that a limited set of redundantly sampled sites with uncorrelated and low observational error (≈0.3¿--0.6 ¿C) will provide the best approach to reconstruction of large-scale features of the SST field from coral data for the preinstrumental period. ¿ 1998 American Geophysical Union