In this paper we establish a foundation for understanding the instrumentation needs of complex dynamic systems if ecological interface design (EID)-based interfaces are to be robust in the face of instrumentation failures. EID-based interfaces often include configural displays which reveal the higher-order properties of complex systems. However, concerns have been expressed that such displays might be misleading when instrumentation is unreliable or unavailable. Rasmussen's abstraction hierarchy (AH) formalism can be extended to include representations of sensors near the functions or properties about which they provide information, resulting in what we call a ''sensorannotated abstraction hierarchy''. Sensor-annotated AHs help the analyst determine the impact of different instrumentation engineering policies on higher-order system information by showing how the data provided from individual sensors propagates within and across levels of abstraction in the AH. The use of sensor-annotated AHs with a configural display is illustrated with a simple water reservoir example. We argue that if EID is to be effectively employed in the design of interfaces for complex systems, then the information needs of the human operator need to be considered at the earliest stages of system development while instrumentation requirements are being formulated. In this way, Rasmussen's AH promotes a formative approach to instrumentation engineering.# 2002 Elsevier Science Ltd. All rights reserved.KEYWORDS: work domain analysis; sensors; instrumentation; abstraction hierarchy; ecological interface design.''What can be sensed forms a fundamental limiting feature of displays. This limiting feature is not always given the emphasis it deserves. [. . .] A measuring instrument which indicated directly and accurately its position relative to the earth's surface would render obsolete the entire science of navigation as we know it. This illustrates clearly the limits imposed on navigation systems by lack of a sensing instrument. The whole system of celestial and radio fixes, the use of compass information, inertial and other ''dead reckoning'' systems, are all substitutes for what we would like to sense directly and display but do not yet have the sensing means for. The initial step in considering the design of the displays for a particular manual control system is to analyze the information the operator would really like to have and to consider the sensing instruments available to obtain it for him. Too often this analysis is not carried out, and it is assumed that the operator requires information that, for instance, has been displayed on similar systems in the past . . . '' Kelley (1968, p. 90-91).