Abstract: Segmentation of MR images was formulated in this thesis as the problem of partitioning a set of feature vectors obtained from an MR image into a relatively small number of clusters. Segmentation of MR images is formulated as an unsupervised vector quantization process, where the local values of different relaxation parameters form the feature vectors which are represented by a relatively small set of prototypes. The complex spatial distribution of the tissue regions, in turn, may cause the MR image intensity in a given pixel to represent signal from a mix of tissues, commonly referred to as the partial volume artifact. In this project a deliver has been done to a MR image for human brain with tumor with a format (jpg) that helps in processed system evaluation (the image was taken from RADIATION AND NUCLEAR MEDICINE CENTER). A method presented to produce segmentation system for multiform tumor disease from trans-axial MR image. The method uses MATLAB 7.0 and applies the algorithm of FCM of c = 14, that is, the segmented images contained fourteen different segments. The segmented image produced by Kohonen's (unlabeled data) LVQ algorithm applied with initial value of the learning rate ?= 0.001 and the total number of iterations was = 20. The tumor and the surrounding edema were clearly identified by all three algorithms from the FALVQ 1, FALVQ 2, and FALVQ 3 families tested in this thesis. The competition between the prototypes during learning and its impact on the performance of FALVQ algorithms were further explored by an additional set of experiments, which evaluated the effect of the free parameter on the performance of various algorithms from the FALVQ 1 family.Combined by applying morphology operation as filter threshold that is, the low threshold was lo= 100 and height threshold hi= 255, and hole and filling by area open instruction of pixels intensity bw= 150. Finally, it is discriminate between normal tissues and abnormalities and obtain image that contain tumor only. From the results it could be found that this segmentation approach is simple and easily implementable, while the use of unsupervised LVQ algorithms does not rely on a priori information provided by human experts.