Existing graphical techniques for data modeling handle diverse user requirements by combining local user views into an integrated schema. This produces a design that does not directly represent the data requirements of any user group in a heterogeneous environment, and which is flexible in the face of changing requirements. This paper offers an explanation of these problems in terms of the treatment of classification in existing data models and diagramming techniques. A new graphical technique is introduced which draws from research on human classification and offers a more flexible approach to modeling multiple views and sharing data among them. Implications of the model for research and practice are also discussed.