A network video playback method and an apparatus, where location information of an independently decoded frame in a media segment and location information of a non-independently decoded frame that is in the media segment and that references the independently decoded frame are added to a segment index in order to determine a video frame closest to an access time point from the independently decoded frame and the non-independently decoded frame. The independently decoded frame that is referenced by the video frame closest to the access time point is acquired when the video frame closest to the access time point is the non-independently decoded frame, and decoding starts from the video frame closest to the access time point according to the referenced independently decoded frame, and video playback is performed. Hence, access accuracy of network video playback can be improved.