Data fusion and sensor management approaches have largely been implemented with centralized and hierarchical architectures. Numerical and statistical methods are the most common data fusion methods found in these systems. Given the proliferation and low cost of processing power, there is now an emphasis on designing distributed and decentralized systems. These systems use analytical/quantitative techniques or qualitative reasoning methods for data fusion. Based on other work by the author, a sensor may be treated as a highly autonomous (decentralized) unit, capable of extracting qualitative behaviors from its data. For example, it detects spikes, disturbances, noise levels, signs, off-limit excursions, step changes, drifts, and other typical measurand trends. In this context, this paper describes a distributed sensor fusion paradigm and theory where each sensor in the system is a highly autonomous sensor (HAS). Hence, leveraging on the rich qualitative information from each HAS, a paradigm and formal definitions are given so that sensors and processes can reason and make decisions at the qualitative level. This approach is conducive to the implementation of intuitive (effective) methods to monitor, diagnose, and compensate processes/systems and their sensors. The paradigm defines a balanced distribution of intelligence (code and/or hardware) to the sensor level, the process/system level, and a higher controller level. The primary application of interest is in intelligent health management of rocket engine test stands. © 2001 by the American Institute of Aeronautics and Astronautics, Inc.