A model of mono- or multimedia document structure is proposed consistent with the notion that there should be few changes in subject between one document passage and adjacent passages, where a passage may be a phrase, sentence, or paragraph. An algorithm for ordering passages in a composition is provided which is optimal given only a concern for the information theoretic similarity of adjacent passages. This procedure is easily coded and works well, given a set of independent subject-related features. This approach is applied to the transformation of linear text into hypertext. An expansion of the model has an economic component which allows for user feedback to modify the structure of the linear or hypertext documents. This model is media independent and may prove useful for analyzing and processing music, images, conversation, and social communication, as well as traditional text. Two figures and four tables illustrate the discussion. (Contains 22 references.) (Author/SLD)