The split window method is successfully being used to retrieve the temperature over sea surfaces from satellite radiances in clear sky and has the great advantage of simplicity. However, such a method does not work over land surfaces, mainly because the emissivity is not equal to 1 and depends on the channel. An extension of this method to apply to land surfaces requires one to take account of emissivity—such an extension is presented in this paper. First, using Lowtran 6, the accuracies of the various linearizations of the radiative transfer equation leading to the split window are checked. This implies that the retrieved surface temperature depends linearly on emissivities and brightness temperatures. Such behaviour has been checked on actual examples. Theoretical equations are then derived which show that the actual surface temperature can again be expressed as a linear combination of the brightness temperatures measured in two adjacent channels with coefficients depending on spectral emissivities but not on atmospheric conditions. Using Lowtran 6 these properties have been verified and the dependence of these coefficients has been explicitly computed leading to a local split window method for the NOAA-9 Advanced Very High Resolution Radiometer. Finally, we show that accurate surface temperatures can be retrieved using this local split window method once emissivities in two adjacent channels are known with sufficient accuracy.