Abstract: Based on the knowledge of the structure of Chinese characters, an efficient method to extract strokes of Chinese characters directly from the original character image is presented in this paper. The theorems of extracting horizontal stroke, vertical stroke, up right slanting stroke, and up left slanting stroke are described quantitatively. And the rules on eliminating stroke noises of various artistic fonts are presented. This method overcomes the difficulties in dealing with the intersection or connection of different strokes, and can eliminate noises successfully. It plays an important role in Chinese character information processing such as character recognition and fonts automation.