This study developed a physical algorithm for retrieving land surface temperatures (LST) by using the special sensor microwave imager (SSM/I) brightness temperature measurements at 19.35 and 22.23 GHz. A nonlinear algebraic equation is obtained for the LST and is solved through the Newtonian iterative procedure. The technique does not require any surface classification because the surface emissivity variations are essentially eliminated by using two closely spaced channels. In addition, the algorithm is less influenced by ice scattering from precipitation and snow cover because of the use of low frequencies. The retrieval accuracy is first examined using simulated measurements. It is found that the root-mean-square (RMS) error is about 3.8 K, with the largest errors occurring in dry atmospheres having total precipitable water of 5.0 mm or less. The RMS error slightly increases to 4.4 K when the algorithm is applied for the actual SSM/I measurements. Large biases were found for cold temperatures where the atmospheric water vapor content may be very low. ¿ 1998 American Geophysical Union