We developed theoretical extensions of Huttenlocher, Hedges, and Duncan's (1991) category-adjustment model of human spatial memory to incorporate the use of fuzzy boundaries and cue-determined prototypes. In two experiments, people reproduced locations of dots in a circle, while the number of external reference cues varied. In Experiment 1, the task field was stable and results were consistent with the use of fixed categories unaffected by number of cues. In Experiment 2, the task field was made dynamic by rotation on most trials, with results evaluated for nonrotation trials. The large cue effects observed for angular bias were consistent with the proposed cue-based fuzzy-boundary model. Large cue effects were also observed for absolute error, consistent with a model in which proximity to cues predicts stability of memory. Results point to the key role of orientation to the task environment in determining whether categorical encoding is based on cues. [ABSTRACT FROM AUTHOR]