This work presents a Computer Vision prototype strategy for the automatic detection of mycotic infections on tomato crops. This Computer Vision method is based on the characterization of tomato leaflets (both healthy and early blight-infected regions of interest - ROIs) by color description (MPEG-7 standard descriptors). A small size ROI collection manually annotated by experts is used for both training and testing of a simple classifier (1-NN). The performance of each descriptor under study (Color Structure Descriptor, CSD; Color Layout descriptor, CLD; and Scalable Color Descriptor, SCD) is analysed by a nested-leave-one-out cross validation. The inner loop permits a individual descriptor configuration evaluation, while the outer loop yields an average performance comparison between different descriptors. Our results show that CSD had a better performance than SCD and CLD.