A novel first-principles-based diagnostic system called PRODIAG is proposed for on-line detection and identification of faulty components during incipient off-normal process conditions. The concepts of qualitative physics reasoning and function-oriented diagnostics are employed in the design of PRODIAG and result in two unique capabilities not found in other plant-level diagnostic systems. First, PRODIAG is fully portable as it requires only modification of the input files containing the appropriate process schematics information to be able to diagnose single-component failures in different processes/plants. Second, PRODIAG detects unanticipated faults. Hence, it does not require the prespecification and formulation of rules to cover every conceivable fault scenario, and unlike traditional approaches, it is not likely to misdiagnose unforeseen events. PRODIAG's approach is to map process symptoms into component faults through a three-step mapping procedure with a knowledge base containing three distinct types of information: qualitative macroscopic balance equation rules, functional classification of process components, and the process piping and instrumentation diagram. The concepts introduced in the proposed diagnostic system are described, and an illustrative example shows how they are used in plant-level diagnostics.