Robustness is one of the important requirements of digital watermark. To improve the robustness of watermarks, some algorithms proposed to apply error correcting coding (ECC) in watermarking, intending to lower the detection error rate by correcting some errors. However, there is an important difference between the watermarking channel and the conventional communication channel. Owing to the constraints of imperceptibility, the redundancy introduced by ECC will result in a decrease of embedding strength for watermarking. In other words, the coding rate of ECC affects the amplitude of embedding signal. Hence, a question arises: For improving watermark robustness, how to choose the coding rate of ECC? In this paper, this problem is addressed both analytically and experimentally. By employing Plotkin Bound and Hamming Bound, the coding rate constraints of Q-symbol linear block codes are derived. The conclusions are demonstrated by the experiment results.