Current sea surface temperature (SST) retrieval algorithms consist of linear equations involving brightness temperatures multiplied by fixed coefficients ai. These coefficients are derived from atmospheric transmission models and radiosonde profiles or from regression against ocean buoy measurements. Errors in SST retrieval occur when the actual atmospheric conditions differ from those assumed when the ai were originally computed; it is generally agreed that water vapor variations are the primary source of this error. We propose a novel radiance-based SST algorithm which compensates for temporal fluctuations in the water vapor profile. The method does not use any form of regression; rather, we utilize a single radiosonde seed combined with full radiative transfer theory applied to an atmospheric model. As well as improving the accuracy of the remotely sensed SST, this approach allows us to estimate the total water vapor content for a given AVHRR pixel. We compare the performance of this technique with two standard SST models, using buoy data from the west coat of Tasmania as surface truth. Preliminary results from a sample of 33 NOAA 9 overpasses gathered over a 10-month period in 1987 are most encouraging: the dynamic water vapor (DMV) algorithm outperforms the standard SST methods in terms of retrieval accuracy, underestimating the buoy temperature by an average of 0.22 K. This result demonstrates that the method has merit and suggests that the total water vapor content returned by the DWV algorithm is probably a reasonable first estimate. ¿ American Geophysical Union 1993