A land surface temperature (LST) retrieval algorithm called week-fixed-parameter method was proposed. The new method is achieved by week-fixed sensitive parameters so as to control information flow from MODIS thermal infrared data. Based on the thermal infrared radiative transfer theory, a forward retrieval model was established and 14 parameters were retrieved together. The retrieved parameters included LST, average temperature of atmosphere, six bands’ atmospheric transmittances and surface emissivities in mid-infrared(3~5 μm) and infrared(8~145 μm). The algorithm was tested with both MODTRAN simulated data and MODIS data. Compared with that of split-window algorithm, MODIS temperature products and ground truth, the RMSE is 0.57 K,which is better than that of traditional split-window algorithm. The test indicates that the weak-fixed-parameter method is effective to retrieve LST.