The information requirement in the health area and the increase difficult for knowledge retrieval inside this domain, demands the existence of tools to support the processing of large amount of existing texts in the information repositories sources. Addressing this problem this work introduces an architecture aiming to support knowledge retrieval process from databases using annotated images in the anatomical domain. The architecture was implemented using five layers (the domain ontology, the annotation, the support, the retrieval and visualization layers) based on the Foundational Model of Anatomy and the Unified Medical Language System, reference ontologies in the biomedical domain. This architecture improves substantially the knowledge retrieval due to the use of images that link its regions with concepts of biomedical ontologies. The architecture proposed can be generalized and applied in other domains employing other reference ontologies. © 2010 ACM.