A split-window equation is derived for land surface temperature, yielding T = T4+ A(T4− T5) + B(ε), whereTis the true surface temperature,T4andT5are the brightness temperatures measured in AVHRR channels 4 and 5,Ais a coefficient related to the atmospheric transmittances in AVHRR channels 4 and 5, being dependent on the atmosphere type and independent on surface emissivity, andB(ε)takes into account the emissivity effect, which depends on both the channel surface emissivities (ε4and ε5) and the atmosphere type. The atmospheric dependence of split-window coefficients,Aand B(ε), is discussed by means of satellite measurements simulations andin situdata. It is shown that linear, global-scale algorithms can produce inaccurate estimates of surface temperature when they are applied to areas of reduced atmospheric variability. A simple, non-linear split-window algorithm has been proposed, in which A = a0+ a1(T4− T5). The constants a0and a1have been calculated from a set ofin situand coincident satellite measurements of sea surface temperatures taken over a wide range of latitudes. A linear split-window algorithm optimized for mid-latitude conditions has been also derived. The combined effects of atmospheric attenuation and surface emissivity on T4− T5have been discussed. Moreover, the theoretical results derived are applied to a NOAA-9 image, for which a procedure for estimating the channel emissivity difference (ε4− ε5) using satellite sensor data is suggested.