Coral skeletons archive past climate variability with unrivaled temporal resolution. However, extraction of accurate temperature information from coral skeletons has been limited by “vital effects,” which confound, and sometimes override, the temperature dependence of geochemical proxies. We present a new approach to coral paleothermometry based on results of abiogenic precipitation experiments interpreted within a framework provided by a quantitative model of the coral biomineralization process. DeCarlo et al. (2015a) investigated temperature and carbonate chemistry controls on abiogenic partitioning of Sr/Ca and U/Ca between aragonite and seawater and modeled the sensitivity of skeletal composition to processes occurring at the site of calcification. The model predicts that temperature can be accurately reconstructed from coral skeleton by combining Sr/Ca and U/Ca ratios into a new proxy, which we refer to hereafter as the Sr-U thermometer. Here we test the model predictions with measured Sr/Ca and U/Ca ratios of 14 Porites sp. corals collected from the tropical Pacific Ocean and the Red Sea, with a subset also analyzed using the boron isotope (δ11B) pH proxy. Observed relationships among Sr/Ca, U/Ca, and δ11B agree with model predictions, indicating that the model accounts for the key features of the coral biomineralization process. By calibrating to instrumental temperature records, we show that Sr-U captures 93% of mean annual temperature variability (26–30°C) and has a standard deviation of prediction of 0.5°C, compared to 1°C using Sr/Ca alone. The Sr-U thermometer may offer significantly improved reliability for reconstructing past ocean temperatures from coral skeletons.