Cross-evaluation of sea surface temperature (SST) algorithms was undertaken using split-window channels of Meteosat Second Generation Spinning Enhanced Visible and Infrared Imager (SEVIRI) as a proxy for the Geostationary Operational Environmental Satellites-R (GOES-R) Advanced Baseline Imager (ABI). The goal of the study was to select the algorithm which provides the highest and the most uniform SST accuracy within the area observed by the geostationary sensor. The previously established algorithms, such as Non-Linear Regression (NLR) and Optimal Estimation (OE) were implemented along with two new algorithms, Incremental Regression (IncR) and Corrected Non-Linear Regression (CNLR), developed within preparations for the GOES-R ABI mission. OE, IncR and CNLR adopt the first guesses for SST and brightness temperatures (BT) and retrieve deviations of SST from the first guess (increments). OE retrieves SST increments with inversion of the radiative transfer model, whereas CNLR and IncR use regression equations. The difference between CNLR and IncR is that CNLR uses NLR coefficients, whereas IncR implies optimization of coefficients specifically for incremental formulation. Accuracy and precision of SST retrievals were evaluated by comparison with drifting buoys. The major observations from this study are as follows: 1) all algorithms adopting first guesses for SST and BTs are capable of improving SST accuracy and precision over NLR; and 2) IncR delivers the highest overall SST precision and the most uniform distributions of regional SST accuracy and precision. This paper also addresses implementation and validation issues such as bias correction in simulated BTs; preserving sensitivity of incremental SST retrievals to true SST variations; and selection of criteria for optimization and validation of incremental algorithms.