The invention discloses a self-adaption dividing method for code units in high-efficiency video coding, which mainly solves the problems that a largest code unit and a smallest code unit in the division of the code units in the prior art can not be regulated according to the characteristics of video content, and the dividing mode quantity of blocks on the boundary of a video frame is few. The method comprises the steps of: firstly, adopting the rate distortion rule or the relativity among video frames to self-adaptively determine the largest code unit and the smallest code unit in each video frame; secondly, expanding a rectangular block within the video frame in the largest code unit on the boundary of a video frame into a square block, and dividing the square block; then, marking the rectangular blocks after dividing, and carrying out frame prediction to the rectangular blocks after dividing; and finally, transforming, quantizing and entropy coding to prediction residuals. The invention has the advantage of high video compression efficiency, and can be applied to high-performance video coding standards.