An approach to chemical plant fault diagnosis is presented that utilizes patterns of violation and satisfaction of the quantitative constraints governing the process. Process knowledge consists of a list of the operational constraints on the plant together with sufficient conditions for violation of each constraint. Interpretation of the pattern of constraint violations is treated by Boolean and non-Boolean techniques. It is shown that non-Boolean reasoning techniques increase the stability and sensitivity of the diagnosis in the presence of noise. The techniques introduced in this paper are easily implemented in rule-based expert systems using certainty factors.