Abstract-The task of decentralized decision-making involves interaction of a set of local decision-makers, each of which operates under limited sensing capabilities and is thus subjected to ambiguity during the process of decision-making. In a prior work [4] we made a key observation that such ambiguities are of differing gradations and presented a framework for inferencing over various local control decisions of varying ambiguity levels to arrive at a global control decision. We develop a similar framework for performing diagnosis in a decentralized setting. For each event-trace executed by a system being monitored, each local diagnoser issues its own diagnosis decision (failure or non-failure or unsure), tagged with a certain ambiguity level (zero being the minimum). A global diagnosis decision is taken to be a "winning" local diagnosis decision, i.e., one with a minimum ambiguity level. The computation of an ambiguity level for a local decision requires an assessment of the self-ambiguities as well as the ambiguities of the others, and an inference based up on such knowledge. In order to characterize the class of systems for which any fault can be detected within a uniformly bounded delay, we introduce the notion of N -inference-diagnosability (for failures), where the index N represents the maximum ambiguity level of any winning local decision. We show that the codiagnosability introduced in [5] is the same as 0-inferencediagnosability; the conditional codiagnosability introduced in [13] is a type of 1-inference-diagnosability; and the class of higher-index inference-diagnosable systems strictly subsumes the class of lower-index ones