Tomato represents an important vegetable crop worldwide. During cropping cycle several diseases and abnormal conditions may affect tomato plants resulting on considerable losses of production. A precise identification of these pathologies in early phases is fundamental for the implementation of control strategies. Nevertheless, the right identification of symptoms of plants diseases require highly specialized knowledge and facilities, which are not available for small growers. Recently, computer vision tools have been proposed as an alternative for tomato diseases characterization. These works mainly focus on identification of affected regions and classification tasks. Nevertheless, non-specialists may lack of clarity about what they are looking for during the assessment. In these cases, Content Based Image Retrieval (CBIR) systems can be helpful as a complementary strategy to improve the quality of the search by allowing exploration of databases with supplementary information. This work presents a novel strategy for image retrieval of tomato leaves for greenhouse crops suitable to support disease diagnosis. The strategy is based on color structure descriptors and nearest neighbors. Experimental results show that the proposed approach can successfully characterize in several abnormal conditions, such as, chlorosis, sooty moulds and early blight.