A procedure for the diagnosis of intermittent faults in combinational circuits is suggested. This procedure employs a probabilistic model for intermittent failures and presumes that a detection experiment has been run. The circuit is assumed to be irredundant and to possess a single fault out of n possible ones. The approach suggested is based on the repeated application of tests that test for these faults had their effect been permanent. A subset of the test set is selected and is repeatedly applied until a failure is observed. Similar subexperiments are then run with appropriate test subsets until the highest diagnostic resolution is obtained. The expected length of the diagnosis experiment is guaranteed to be finite. This is shown by proving that the expected length of each subexperiment is finite. The diagnosis experiment can be terminated, when any preset time limit is exceeded, compromising the obtained diagnostic resolution. Local symmetry of the fault table is found to be the necessary and sufficient condition for maximum diagnostic resolution.