This paper proposes a framework for the retrieval of mul-tiple regions characterized by low-level features. The re-trieval combines the assessment of the visual similarity between regions and of the similarity of the relationship between these regions. Binary Partition Trees (BPTs) are used as a basis of the image representation. Regions of the BPT are described by low-level descriptors. Finally, relevance feedback is used to avoid the need of manually setting the weights associated to each descriptor.