Automatic detection of retinal image exudation is helpful for early diagnosis of diabetic retinopathy. According to the decision tree theory, we propose a retinal intrusion detection method based on color fundus images and test it with Messidor database. It can help us to distinguish the diseased images and normal fundus images. The experimental results show that this method is better than the eye specialist's manual judgment in that it can detect exudation area more accurately in a changing enviroment of light by the integration of images.