The presenters will describe a technique for creating gridded composites of sea surface temperature (SST) from the MODIS Aqua and Terra satellites. Several features of the technique make it well suited to numerical weather prediction. First, it can be applied in real time. Second, the temporal component of the technique can be adjusted to maximize autocorrelation in the data and minimize missing data where cloudiness and other circumstances make good retrievals impossible. Third, holes in the composite owing to any remaining missing data are filled with a background field. Experience demonstrates that the SSTs in MODIS composite differ by more than 5-10 K from the SST fields that are commonly used to define the lower boundary in numerical simulations.