This paper develops a data reduction framework for assigning fuzzy membership values to continuous geographic data. The goal of classification is defined quantitatively using explicit criteria of error and confusion introduced by the classification process. A new method of assigning fuzzy membership values is designed to reduce overall error, and compared with standard, similarity-based methods. As a case study, a continuous forest-type map is created for an area of the northeastern United States using data from the U.S. Forest Service. Given certain reasonable assumptions regarding the interpretation of continuous classes, the new method is shown to provide small but consistent reductions in error and confusion. More generally, the data reduction framework provides explicit meaning to the use of continuous classes in ecological mapping, allowing for quantitative measurement of the error introduced by the classification process.