Detecting of abnormalities of magnetic resonance image (MRI) of human brain is achieved in this thesis by applying a tumor segmentation method.Recently, many searches are interested with the segmentation of MR image into different classes that are regarded as the best available representation for biological tissues. Mainly, the aim of this project is to design and test an operating system so as to carryout segmentation of MR image of human brain with a tumor. This system is based on fuzzy clustering and knowledge based analysis. The above segmentation system consists of six sequential processing stages as follows : implementation of fuzzy c-means (FCM) algorithm, extra cranial tissues removal, retaining classes containing cerebro spinal fluid (CSF) and pathology, necrosis removal, ventricles removal, and finally, removal of non-tumor spread pixels in the resultant image to obtain image containing tumor only. The first stage of the proposed segmentation system is to starts with implementing a fuzzy c-means (FCM) algorithm to obtain initial segmentation of MR image into ten clusters. The second stage involves the removal of extra cranial tissues using different image processing techniques such as morphological operations. The remaining tissues are reclustered into seven clusters. In the third stage of segmentation system, classes containing CSF and pathology are retained by storing them in a binary image. The other remained classes (consisting mostly of gray matter (GM) white matter (WM) tissues) are also stored in a second binary image helping for ventricles removal during the next stages. Removal of any necrosis from a binary CSF/pathology image is achieved in the fourth stage. This is done throughout smoothing of CSF/ pathology image using Gaussian filter. Then, the smoothed image is processed using T1-histogram thresholding technique to obtain image without necrosis. The ventricles removal is involved in the fifth stage at which the formation of a polygon approximation is achieved. The last intersect (after filling) with the image obtained after necrosis removal followed by removing of the common pixels. Finally, the spread pixels and the clear edges of the image obtained from the previous stage are removed using series of morphological operations and filtering process to obtain image that contains tumor only. An image with tumor only gives some details about the segmented tumor such as the size, perimeter, and flatness. Then an image with tumor only is used to perform recognition process to classify input MR images as normal or abnormal. As a final step of this work, a description was introduced for different images that were obtained starting from an input image reaching to image with tumor only. This description included both of shape and istogram features for each input image.