Many different data modelling or representation schemes have been used or proposed. One important use of such data representations is to communicate the data content of a proposed system design to users: the "user validation" task. The effects of the characteristics of four such data models on user comprehension were investigated in a controlled laboratory experiment. The results showed that the two primarily graphical representations were more understandable than two alternatives for most of a set of tasks designed to simulate user validation. There were some preliminary indications that the graphical or "semantic" data models led to more systematic data modelling behavior. Relational models did out-perform graphical models with respect to relationship identifier recognition. Additional research is discussed that will more fully explore the role of data representations in systems development. The results of this experiment are also be applicable to "end-user computing."