An adaptively imperceptible video watermarking algorithm using entropy model is proposed. This algorithm works by first combining the Human Visual System (HVS) with block-matching techniques to obtain motion-related information. It then uses entropy model to statistically analyze the obtained information, and eventually establish a set of non-linear formulas based on HVS and motion information. Based on the contents of video frames, this set of non-linear formulas can adaptively calculate the maximum strength of every block. Experimental results show that using entropy model and non-linear formulas can significantly improve watermarking imperceptibility, effectively resist common attacks for video watermarking, and consequently achieve higher robustness.