Digital watermarking is a new method for network traffic tracing. Existing watermarking schemes have an obvious phenomenon that embedded watermarks have nothing to do with flows themselves. When we trace multiple flows, the watermark need to be encoded again which will reduce the capacity of watermarks greatly. Therefore, we propose an Entropy and Hash based Double Redundancy (EHDR) watermark model for multi-flow tracing. By using entropy and hash for feature extraction, data flows can be labeled. With the idea of double redundancy, EHDR model not only can reduce the overhead of time and space during the watermark detection, but it also can improve the detection efficiency effectively with the same redundancy. Moreover, this model has a good portability, to be used in many other watermarking schemes. In order to validate the efficiency of EHDR model, we introduce the real traffic from the Center for Applied Internet Data Analysis (CAIDA) dataset into the simulation environment. Experimental results show that this model is able to track the multiple flows, and improve effectively the robustness of the original watermarking scheme as well.