Detecting the occurrence of clear-cutting is a practical use for remotely sensed data, and the image differencing method is one technique that can be used. This method is simple and powerful, but a systematic evaluation of its accuracy and a statistical determination of threshold values have not been established. We examined the accuracy of image differencing methods using a single band or band combinations, or scene combinations, to detect clear-cuts with LANDSAT/ TM data. Furthermore, we examined the relationship between clear-cut size and detection accuracy with a focus on the most effective bands. Results showed that bands 3 and 7 are optimal because they detect clear-cut areas with less error. Accuracy varied with the clear-cut size, and clear-cut regions less than 1 ha in area exhibited more error.