A method to solve the problem of dim target detection under complex background based on their multi-scale fractal feature is presented in this paper. It is realized by calculating the difference of fractal features between natural background and man-made target in variation with scale to extract the MFFK parameters, a multi-scale fractal feature distribution image can be obtained. After that, self-adaptive threshold segmentation method is used on the calculated feature distribution image to detect the dim target. Experimental results show that the proposed method not only suppresses the interference of the complex background effectively, but also presents excellent detection property and environmental adaptability.