A fast loss recovery scheme for IP multicast services is presented in this paper. In this scheme, the head field of retransmission request packets contains the one-way delay between the requestor and the sender of the lost packet, and the head field of repair packets contains the one-way delay between the repairer and the requestor. Each member in the multicast group compares its own one-way delay to the sender or the requestor with those advertised in the request or repair packets, and decides how long it should wait before sending a request or repair packet in the next recovery round. This distributed request/repair timer setting algorithm shows significant performance gains in terms of loss recovery latency and the number of duplicated requests and repairs, compared with the SRM scheme and its adaptive version. Being a pure application-layer scheme, it can be deployed quickly over existing IP multicast infrastructures without extra router assistance. The ability to recover from errors in typical one round-trip time makes this scheme a good candidate for loss recovery in real-time multimedia multicast.