Anonymous network is widely used to access the Internet, causing varieties of cyber security incidents because of its anonymity, which increasingly affects the security of cyberspace. How to detect anonymous network flow to position the anonymous users, is becoming to a research hotspot. However, with rapid development of the encryption and network technology, it is a nontrivial task to detect and position the anonymous user in such a complex network environment.In this paper, we design a prototype system called Watermark based Tor Cross-domain Tracking System that is effectively detects and determine the sender and the receiver on the real Tor network to testify its function. Moreover, instead of conventional passive network flow analysis, this paper learns from active network flow analysis to design three digital watermark models to implement the embedding, extracting and matching of watermark information, and meanwhile it will not affect the network flow's content and transmission. Experimental results on the real data sets show that when embedding the three watermark models on the sender, watermark based Tor cross-domain tracking system indeed yields the positioning function.