85%。 In order to provide parameters for the motion of cotton fruit auto-picking manipulator, a new cotton recognition method was proposed. The method can identify cotton fruits from surroundings correctly, acquire their precise location information, and accordingly pick up automatically. By analysis on the color data of different parts of cotton in RGB space, a conclusion is obtained that the values of red and blue color of cotton fruits are distributed on the line of 45°, while those of cotton leaves and stems deviate from this line. Base on this, the method using color subtraction information is designed. Furthermore, in order to improve the accuracy of cotton recognition, according to shape properties, a new dynamic edge tracing method via Freeman chain coding is used to remove the background noise. Experimental results show that the proposed method has good performance of identification, and the accuracy is more than 85%.