The embodiment of the invention provides a method and device for detecting faults. The method comprises the following steps that: a test point sends a loop-back command to a net element in a service path, wherein the net element receiving the loop-back command performs loop-back on service data at an inlet port or an outlet port thereof; and the test point affirms the specific positions of fault points according to whether the received service date after the loop-back is normal, and if the specific positions of the fault points cannot be affirmed, the test point continues to send the loop-back command to the net element in the service path until the positions of the fault points are positioned according to the receiving condition of the loop-back service data. By applying the method and the device, the positions of the fault points in the network can be accurately positioned.