Fires are one of the major hazards to the forest. Governments of various countries spend an enormous amount of manpower and financial resources to prevent forest fires every year. Many research projects, which monitor and identify the forest fires, have been developed by using infrared remote sensing. In this paper image features of all kinds of forest fire and background surface objects on the visible light image are analyzed. On the base of spectrum feature of the fire and background surface objects in the visible spectrum, feasibility of the method using visible light image to recognize forest fire is discussed, and then a new method with visible light image is put forward to identify forest fires. Firstly, the features of serial forest fire frame are generalized, which establishes the theoretical foundation for further recognition. Secondly, suspicious fire area is found in single frame according to the difference of fire and background got from the first step. Thirdly, some erroneous judgments are eliminated in response to continuous serial frames. In the end Jinggang Mountain which locates in Jiangxi province of China is taken as an example. More than 30 segments of forest fire image flow, photographing in different climate and from various angles are used to identify forest fire. Conclusion can be drawn that the rate of recognition is 100%, and the rate of accuracy is 85%.