A major limitation to brain imaging studies is experimental noise generated at multiple levels, including the measurement instrument, the individual's responses to a task and anatomical differences that make precise localization difficult. To accommodate these limitations, groups of test subjects are necessary, and anatomical variations among individual brain MRI's are reduced through a geometric \"warping\" technique that transforms each individual's brain to a standard norm, upon which functional activation patterns are superimposed. In this way, reliability of the data obtained is increased. The disadvantage of such a process is the need to ignore individual differences in the absence of proof that they do not actually constitute noise. A method in which an individual's activation patterns can be reliably superimposed upon their own anatomy prior to being entered into the data analysis would reduce at least one source of error. Because subtle anatomical variations exist even among normals, yet major structures have remarkable consistency between persons, we have begun to explore the potential of establishing fuzzy \"maps\" by which an individual's activation pattern can be tied directly to their anatomy in an efficient, unbiased way. We have already demonstrated the feasibility of this approach with a fuzzy rule based method to identify the frontal lobes, in which the major landmarks that an expert uses to identify brain structures were translated into rules and identification accomplished when multiple criteria were applied. We have determined that in order to extend this method to more complex structures, additional data needs to be incorporated in a flexible model. An example of a complex structure of interest is the hippocampus, a curved structure without clear identifiable borders. We have turned to the anatomical literature in which individual differences in brain structure are often extensively catalogued, most commonly to aid the surgeon, and in which the range of \"normal\" variation is presented in statistical form. Our intention is to use membership values derived from the incidence of anatomical variation in the normal population and to assign fuzzy values to borders where landmarks are less distinct. We will determine whether such a process can generate an anatomical segmentation suitable for functional activation studies. Such a process should obviate the need for the extensive data reduction that inevitably occurs in conventional warping technique. The variety of data available and the strategies tested thus far will be presented and discussed.