This paper builds on research into using multimedia and hypermedia as creative writing tools and reports on a study in using a hypermedia authoring program with middle-years students at an Australian secondary school. The study explored a classroom where collaborative work built on the technical facilities and expertise of young people. The paper presented here focuses on the authors' development of an assessment matrix for evaluating hypermedia stories. The aim is to provide an effective feedback mechanism on the hypermedia work and to give teachers guidelines for planning lessons, and for making summative as well as formative judgments. Four, equally important dimensions are identified as the key components of hypermedia: image, language, interactivity and structural design. Image relates to how the screen looks: the background and foreground, positioning, colour, animation and source of images (e.g. clipart, photographs, or the use of hand-drawing). Language involves written language as signs, or captions within the created hypermedia world, or as attached commentary. Language may also be audio-files containing conversations, commentary, or even chanting or singing. Interactivity refers to the way the viewer or the interactor comes into a relationship with the hypermedia. This involves the interactive nature of the screen, that is, what can be opened up and explored within each screen as well as how the viewer/interactor moves between screens. Structural Design relates to the way the story fits together image, language and interactivity, and the way one aspect relates to another in terms of complexity and cohesion. In order to systematically assess the qualitative variation across the range of students' hypermedia stories, each of the four aspects highlighted above was also considered in terms of the five-level "SOLO" taxonomy (structure of learning outcomes). This taxonomy includes the following five levels: (1) Pre-structural--where the response is really not relevant to the question; (2) Uni-structural--where an appropriate but minimalist response is given; (3) Multi-structural-- where several responses (or answers) are given and the responses are appropriate, but there is no relationship between the responses; (4) Relational--where all responses are related into a reasonably coherent argument; and (5) Extended abstract--where responses come together in an argument, which goes beyond any of the information previously given or hinted at. When the four dimensions mentioned above are mapped against the five qualitatively different levels of response, a series of taxonomies is produced, which gives a clear indication of the strengths within each story, as well as the variation in quality across all the stories. In so far as this method maps and explores qualitative variation within a specific phenomenon it can be seen as phenomenographic. In this paper, three of the hypermedia stories' matrices are explored and one story is assessed using the whole method.