In this paper, we use multifractal spectrum (MFS) based on sum measure, minimum measure and maximum measure to diagnose and segment images of rapeseed leaf eroded by rapeseed sclerotinia, rapeseed leukoplakia and rapeseed phytomyza atricornis, respectively. The results illustrated that the image segmentation obtained by the MFS method are superior to those obtained by traditional threshold method, C-means clustering method and region growing method. The advantage of the MFS method is mainly manifested in segmenting the disease leaf edge clearly and locating the lesion area accurately while maintaining more details with strong locality and high accuracy. By comparison, the segmentation results obtained by the max and min capacity measure are superior to those by the sum capacity measure.