Abstract: In this paper,a rough sets based global reduction approach,which is suitable for imaging spectrometer image is proposed.According to the fuzzy sets and rough sets theory,the set of original bands is first automatically divided into several approximative equivalent intervals. In each interval, equivalent bands are linearly combined into one by using genetic algorithm under the consideration of the combined images of other equivalent intervals.Experimental results indicate that the global reduction approach is not only relative faster, but also better than independent reduction approaches.Taking the global reduction as the preprocessing step of imaging spectrometer images, we can apply traditional remote sensing image classification approaches to this type of remote sensing image data.