The accurate identification of the brain tumor boundary and its components is crucial for their effective treatment, but is rendered challenging due to the large variations in tumor size, shape and location, and the inherent inhomogeneity, presence of edema, and infiltration into surrounding tissue. Most of the existing tumor segmentation methods use supervised or unsupervised tissue classification based on the conventional T1 and/or T2 enhanced images and show promising results in differentiating tumor and normal tissues [1-3]. However, perhaps due to the lack of enough MR modalities that could provide a more distinctive appearance signature of each tissue type, these methods have difficulty in differentiating tumor components (enhancing or non-enhancing) and edema. These issues are alleviated by the framework proposed in this paper, that incorporates multi-modal MR images, including the conventional structural MR images and the diffusion tensor imaging (DTI) related maps to create tumor tissue profiles that provide better differentiation between tumor components, edema, and normal tissue types. Tissue profiles are created using pattern classification techniques that learn the multimodal appearance signature of each tissue type by training on expert identified training samples from several patients. The novel use of DTI in the multi-modality framework, helps incorporate the information that tumors grow along white matter tracts [4]. In addition to distinguishing between enhancing and non-enhancing tumors, our framework is also able to identify edema as a separate class, contributing to the solution of tumor boundary detection problem. Tumor segmentation and probabilistic tissue maps generated as a result of applying the classifiers on a new patient reflect the subtle characterizations of tumors and surrounding tissues, and thus could be used to aid tumor diagnosis, tumor boundary identification and tumor surgery planning.