The European Space Agency (ESA) is presently supporting the Processes Research by Imaging Space Mission (PRISM), a project whose aim is the accurate retrieval of land surface temperature (LST) using a new thermal radiometer. Two previous papers have already addressed some aspects of the design and performance requirements for this thermal sensor: the analysis of different algorithms for emissivity-temperature decoupling and the techniques for atmospheric and emissivity correction. This paper focuses on the selection of the most appropriate spectral channels for the PRISM thermal instrument. For this purpose, a theoretical argument was first used to obtain a preliminary selection of potential spectral channels in the 8.0--12.5 μm region. A simulation analysis, which contained a global database made up of 6120 surface-atmosphere conditions, was subsequently performed to determine the LST retrieval error for each of the preliminary combinations of channels. Results indicate that channel configurations consisting of two separate bands in the 10.0--12.5 μm atmospheric window are the most suitable for LST retrieval. However, there is no single pair of optimal bands: there are 22 possible optimal band configurations which produce the minimum error in LST estimates (¿0.9 K). Nevertheless, an alternative configuration was also found: the 8.5--9.0 μm band or the 8.6--9.1 μm band as channel 1 combined with the 8.0--8.5 μm band as channel 2 enable a LST retrieval with an error of ¿1.1 K. ¿ 1998 American Geophysical Union