When missing, ambiguous or distorted data is available in digital image processing,soft computing (e.g. fuzzy logic, neural networks, evolutionary computation) has provedto yield promising results. In the field of biomedical image analysis, when information has astrong structural character, many methods of artificial intelligence are well suited for knowledgediscovery, representation and processing. Fuzzy logic acts as an unified framework forrepresenting and processing both numerical and symbolic information, as well as structuralinformation constituted mainly by spatial relationships in biomedical imaging. This paper describesthe use of fuzzy logic at low level to a higher level (e.g. model based structural patternrecognition and scene understanding). Applications are for segmentation of brain structuresin magnetic resonance (MR) and CT (computer tomography) images, based both on atlas andreal data. Promising results show the superiority of this knowledge-based approach over besttraditional techniques in terms of segmentation errors.