Abstract: To segment automatically a cerebral RMN image in a reliable and robust manner is a delicate problem. However, it is a task that the expert could do with a certain precision, if he had one unlimited time. The expert has a very precise priori knowledge of what he wishes to segment. He knows the general shape and the disposition of the objects that he must segment. He can use principles logical of "common sense". This type of knowledge is integrated, to different degrees and often of very implicit manner, in all methods of segmentation. These methods are generally founded on very different principles as for example the processes of classification, the use of models of deformable contours or models of knowledge. But these last don`t use in general that only one volume of image what causes in most cases a considerable loss of information. Besides these methods tend to be dependent of some parameters as the number of classes, the functions of similarities that are often imposed what implies that the application of these techniques becomes non suitable to the multimodal medical images. The goal of our work is to show that it is possible to define a common setting of work to permit the implementation of cooperation between heterogeneous approaches. The interest of such an approach is to be able to exploit the complementarity of information that results from the application of several methods in order to propose a complete system of segmentation. For it we developed one automatic classification algorithm baptized ONFPCM (Optimal Neuro-Fuzzy Possibilist C-Means) using a supplementary stage of Bootstrap by the slant of an auto-organizing neuron network algorithm named CENN (Capture Effect Neural Network). In this study a number of classic automatic classification methods as the HCM algorithm (Hard C-Means) and FCM (Fuzzy C-Means) are elaborated as well as the NFMEP algorithm (Neuro-Fuzzy Maximum Entropy Principle). These last are applied on multimodal medical images and the results are compared to those gotten by the ONFPCM algorithm.