Abstract: Multi-scale structural similarity (MS-SSIM) is a commonly used full-reference quality assessment metric. Since the original video is required for reference, it is not suitable to be applied in real-time assessment of network video quality. In this paper, a no-reference MS-SSIM video quality assessment model based on H.264 bitstream was proposed. First, I-frame and P-frame encoding mode and motion vector parameters from H.264 bitstream were extracted from H.264 bitstream, which then were statistically analyzed to characterize the richness of the texture and the intensity and complexity of the motion of the video. Second, the parameters were combined with quantization parameter to form the bitstream feature parameter set, and support vector regression (SVR) was finally applied to the relationship between the bitstream feature parameters and MS-SSIM to predict the video quality metric of MS-SSIM for H.264 bitstream. The proposed model only uses the parameters extracted from the video bitstream and does not demand to decode the video bitstream completely. Compared with a state-of-the-art no-reference bitstream prediction model, the proposed model can achieve higher prediction accuracy.