Abstract: So far there is no theoretically effective method for the selection of optimum spectral channels for multispectral remote sensing. In this paper a new method is proposed which is based on the criteria of maximum information. As strong correlation exists among signals in different spectral channels, this factor must ibe seriously considered during the computation of amount of information. Prom the measured spectral characteristics of main ground objects, mutual correlation factors of all spectral channels are computed. If signals are supposed to be normally distributed, then by computing determinants of correlation matrices, conditional information of each channel signal can be obtained. By comparing the total amount of information of every possible combination of channels, the optimum combination can be found. When the number of possible channels is too large, the computation process may be very complex and time comsuming. Asimple method is also recommended which can give "quasi-optimum" selection of spectral channels. An example is shown for illustration, and the result is compared with the experimental result in foreign literature. Some comments are made by the author for the design of future remote sensing satellite mutisipectral systems.