A multiple-band algorithm is proposed in this paper to separate land surface temperature and emissivity from ASTER data. Three methods can be used to solve the equations. The first is the performance of classification for the images and the formulation of  different equations, followed by the solution of the equations. The second is least-squares. The third is the simulation of the database according to the characteristics of object emissivities and the utilization of the neural network to solve equations. An analysis indicates that the neural network can improve the practicability and accuracy of the algorithm. The accuracy of neural network proves to be very high for the test data simulated from MODTRAN 4. An application example is given in this paper, and the analysis suggests that the neural network also possesses the self-study capability. The simulation data show that the average error of land surface temperature is below 0.5℃, and the error of emissivity in band 11~14 is below 0.007(band 11，12)and 0.006 (band 13，14), respectively.