While a multiple-path interconnection network is capable of tolerating any single fault, the knowledge of the fault's location is required before it can adapt itself. An approach of online single-fault detection is given. Based on a new fault model, a systemwide diagnostic procedure is developed to effectively detect and locate a single fault throughout a fault-tolerant network. The model is realistic and has potential usefulness as a tool for modeling faulty states of larger switching elements (e. g. , n multiplied by n switching elements, n less than 2). Networks under diagnosis behave in a distributed-control manner, i. e. , a tag needed for establishing a path is conveyed by the same resources (switching elements and links) as those for transmitting data. Test vectors for appropriately setting switching elements when the procedure is conducted are presented. Faults are classified into two different groups each of which is dealt with separately to ease the diagnostic procedure.