Reef-building corals provide seasonally resolved records of past climate variability from the ocean via variations in their oxygen isotope composition (δ18O). However, a variety of non-climatic factors can influence coral δ18O including processes associated with coral biomineralization and post-depositional alteration of the coral skeleton, which add uncertainty to coral based paleoclimate reconstructions. These uncertainties are especially large in mean climate reconstructions developed from coral δ18O values due to the large variability that exists in mean skeletal δ18O signatures. We present a novel framework to minimize this uncertainty in mean coral δ18O records based on a regression model that uses four commonly measured properties in coral skeletons and associated coral δ18O records. We test the ability of the model to reduce noise in a Holocene climate reconstruction comprised of 37 coral δ18O records from Kiritimati in the equatorial Pacific. Up to 43% of the variance in the detrended Holocene dataset is accounted for by a combination of four predictors: (1) mm-scale variability in a coral δ18O record, (2) the physical extent of diagenetic alteration, (3) coral extension rate, and (4) the mean coral δ13C value. Once these non-climatic artifacts are removed from the reconstruction, the weighted variance of the Holocene dataset is reduced by 46% and the uncertainty in the trend of coral δ18O over time is reduced by 26%. These results have important implications for the climate interpretation of this Holocene data set. This framework has the potential to improve other paleoclimate records based on ensembles of coral δ18O records.