This paper considers problems inherent in constructing typologies based on subdividing artifacts using clustering algorithms that assume a paradigmatic structure (as do most, if not all, currently used algorithms) and presents an alternative method based on a taxonomic, rather than a paradigmatic, structure for the artifact classes. The method is exemplified by analysis of an assemblage of utilized flakes from a series of late prehistoric habitation sites in the central highlands of Peru. These flakes were previously analyzed using more intuitive, traditional methods but without identifying any satisfactory groupings within the assemblage. Similarly, paradigmatically based clustering algorithms applied to the entire data set and using the several variables measured over these flakes were also unable to extract meaningful groupings. In contrast, the method presented here was able to (1) construct a partition of the flake assemblage into groups that also made intuitive sense to the practitioner, (2) have functional interpretation assignable through a model of transmission of forces from agent to object acted on, and (3) produce units differentially occurring across the several subsites from which the utilized flakes were obtained, i.e., units that add another dimension to the analysis of activities represented by the archaeological assemblage. Although demonstrated with utilized flakes, the method has general application.