As a basis for an automated interpretation of magnetic resonance images, the paper proposes a fuzzy segmentation method. The method uses five standard fuzzy membership functions: small, small medium, medium, large medium, and large. The method fits these to the modes of interest in the image histogram by means of a piece-wise linear transformation. A test example is given concerning a human head image, including a sensitivity analysis based on the fuzzy area measure. The method provides a rule based interface to the physician.