In order to improve the performance of the packet- layer quality assessment model of networked video,a frame type detection method without any help of video decoding is proposed only by using the information extracted from the packet header of the video stream.In this method,first,the statistical law of the compressed data of each type of frame is obtained according to the characteristics of different types of frames.Next,the dynamic threshold method is employed to preliminarily estimate the type of each frame in a sliding window.Then,the preliminary estimation results are modified based on the periodicity of group of pictures.Finally,the prediction structures of B- frames are determined by using the Spearman’ s rank correlation coefficient.Experimental results show that,as compared with the method based only on dynamic thresholds,the proposed method achieves a detection accuracy increment of about 20.04%.