Network intruders try to hide their identity by relaying their traf c through a number of intermediate hosts, called stepping stones. Network ow watermarks have been used to detect such attacks by inserting a special timing pattern into one ow by means of arti cial delays and detecting relayed ows by searching for the same pattern. We study the application of coding schemes to improve the ef ciency of network ow watermarks. In particular, we use the Repeat-Accumulate codes, a class of low complexity, high performance error-correcting codes, to improve the detection performance of a recent ow watermark, the RAINBOW. We show the effectiveness of the improved scheme, CRAINBOW, through simulation and discuss design tradeoffs.