The purpose of this study was to validate our MR tissue segmentation technique using a hamster brain tumor model and malignant brain tumors in man. We used a multispectral tissue segmentation analysis. Three sets of MRI data were included: proton density, T2-weighted fast spin echo, and T1-weighted spin echo, as inputs. Three image preprocessing steps included correcting image nonuniformity, application of an anisotropic diffusion type filter, and data point selection by a qualified observer. We used the k-Nearest Neighbor segmentation algorithm, which does not require prior knowledge of the sample distribution. This choice allowed us to optimize the different tissue clusters present in three-dimensional (3D) feature space. In vivo validation of the technique was performed in hamsters harboring tumors induced with JC virus-transformed HJC-15 cells, as compared to three control animals. Human brain tumors obtained by stereotactically guided biopsy in six patients were also included in the study. Finally, brain tumors were removed from two patients who underwent conventional craniotomy using segmentation-derived images as a guide. In the hamsters, 10 tissues were correctly identified by segmentation and were confirmed histologically (P < .02). In the patients, there was also a strong correlation between our segmentation results and the tissue obtained by stereotactic biopsy (P < .01). In one of the two patients who underwent open craniotomy, segmentation images were useful in revealing tumor spread into vital areas of the brain (motor area). In conclusion, the results of segmentation correlate well with the tissues in vivo and thus warrant further clinical utilization and evaluation.