We address, in this paper, the problem of diagnosing intermittent sensor faults. In order to do so,we employ a model of intermittent loss of observations recently proposed in the literature, and use this model, together with an appropriately modified label automaton, to change the problem of detecting intermittent sensor faults into a problem of diagnosing the language generated by an automaton in the presence of intermittent faults, where the fault event is the unobservable event that models the non observation of the event whose occurrence is recorded by the sensor subject to intermittent fault. We present necessary and sufficient conditions for diagnosability of intermittent sensor faults and propose two tests to verify intermittent sensor fault diagnosability: the first one based on diagnosers, which can also be used for online diagnosis, and a second one, based on verifiers, which has the advantage of having polynomial time complexity. (C) 2017 Elsevier Ltd. All rights reserved.