The determination of land surface temperature (Ts), through satellite images, has been object of several studies. The product MOD11A1, broadly used by scientific community, supplies good results for Ts with a spatial resolution of 1 x 1 km. Ts supplied by MOD11A1 is obtained from radiance image using a split-window that depends on emissivity, brightness temperature (Tb), precipitable water and zenith angle, and other empiric parameters. Although the results are reliable, it is observed some incoherence between supplied Ts and the variation of the main parameters involved in the method, for example: the emissivity and Tb. In this work we present an efficiency form of the split-window presented in Kidder & Vonder Haar (1995). The land surface temperature can be estimate by a direct relationship with Tb. That relationship is established by the atmosphere transmittance. So, estimating Tb and the atmosphere transmittance with good accuracy one obtains good results with Ts estimation. The atmosphere transmittance is always attributed to the water vapor. Then a radiative transfer code must be used for such task. In this work, it is used the parameterizations of Robert (1976) applied to atmospheric profiles of MOD07 product. Tb is estimated by the inversion of Planck's Law, where the mean wavelength for each band, takes into account the sensor spectral response. The method is tested with Ts from Becker & Li (1990)'s method and Ts by MOD11A1, using a radiance MOD021km image. The validation place is Paraíba State, a semi-arid region of Northeast of Brazil. Over 379 observed pixels, the Ts mean value for the presented method is 0.3 K larger than the average values supplied by MODA11, while the Becker & Li method provides a 1.7 K for the same set of pixels. Other images together with surface measurements during the satellite overpass it is necessary for final conclusions.