Fuzzy modelling provides methods and techniques for qualifying and quantifying imprecise and uncertain information. The main advantages of fuzzy design representation are fluency, abstraction and continuity, at a level similar to that of analogue techniques, as well as the possibility of local autonomy, i.e. segmentation of a representation into self-regulating and cooperating components. The paper investigates the applicability of fuzziness to digital architectural sketching of floor plan layouts. Based on an analysis of the paradigmatic dimension in analogue floor plan sketches three alternative forms are proposed: (1) Canonical objects with tolerances, (2) objects described by minimal and maximal values, and (3) point sets which decompose the form of an object into a number of discrete, autonomous particles that describe the object by their position and spatial or structural relationships.