Abstract.With the increased utilization of cognitive models for designing user interfaces several disciplines started to contribute to acquiring and representing knowledge about users, artifacts, and tasks. Although a wealth of studies already exists on modeling mental processes, and although the goals of cognitive engineering have becom e quite clear over the last decade, essential epistemological and methodological issues in the context of developing user interfaces have remained untouched. However, recent challenging tasks, namely designing information spaces for distributed user comm unities, have led to a revival of well known problem s concerning the representation of knowledge and related issues, such as abstraction, navigation through information spaces, and visualization of abstract knowledge. All of these issues are associated with mental processes and thus, might becom e part of cognitive models. In this paper we reveal epistemological and methodological assumptions in the ® eld of cognitive modeling as well as their implications for user interface design. It turns out that in order to achieve the goal of developing hum an-oriented (in contrast to technology-driven) hum an-computer interfaces developers have to develop knowledge of the structure and the representational dynam ics of the cognitive systems which are interacting with the computer. We show that in a ® rst step it is necessary to study and investigate the diOE erent levels and forms of representation that are involved in the interaction processes between computers and human cognitive systems. We propose a hybrid user modeling approach as part of the task-based development procedure in TADEUS (Task Analysis/Design/End User Systems). The hybrid approach does not only enable the representation of functional roles end users have to perform, but also how end users perform these roles, i.e. the representation and re¯ection, if not prediction of their behavior. This way, holistic system developm ent that equally takes into account the organizational requirem ents and the end user reality at work places is facilitated.