The invention discloses a segmentation method of news program, which comprises the following steps of: acquiring the audio and video data, selecting new clips about a moderator, an interviewer and program music, training a global background model according to clips data, respectively self-adapting the speech/music clips on the global background model, computing a mapping vector as a space vector to obtain a self-adaptive model, segmenting the news program according to the end point detection, adjusting a threshold value, performing the speech/music recognition one segment by one segment according to the self-adaptive model, labeling the identity of a producer of each segment of speech/music; meanwhile, preserving the confidence grade information, extracting a video key frame from the news program clips with the confidence grade less than a predetermined value, verifying a labeling result, and combining the news program clips which are adjacent to each other and have the same producer of speech/music. By adopting the technical scheme, the automation of the segmentation of the news program is greatly increased, and the accuracy of the segmentation of the news program is guaranteed.