The land surface temperature is an important parameter of the physical process of interaction between the earth and the atmosphere and is widely applied to the research of hydrology, meteorology, global carbon balance and global change. The relationship between the radiation brightness of two thermal infrared channels is directly established to replace the simplified Planck function for solving a radiation transmission equation set, an atmospheric parameter (the specific value of atmospheric downward radiation to atmospheric upward radiation) is also introduced to enhance the accuracy of the simplified radiation transmission equation, a split window algorithm based on the radiation brightness (RBSWA) is put forward and the algorithm is applied to MODIS data land surface temperature retrieval. Compared with the conventional split window algorithm based on the brightness temperature, the land surface temperature retrieval accuracy of the RBSWA is enhanced for about 0.5K. The principle of the method is also suitable for other thermal infrared remote sensing data.