AbstractThe DPCM, a representative prediction encoding system of picture signals, produces two conflicting noises‐overload and granular. This paper describes the selective DPCM that separates picture signals into edge and slant components, and an optimum encoding is applied for each component to consider the two noises. First, statistical analyses on picture signals conclude that they are composed of the edge component and slant component. Second, the edge component, which shows frames of pictures, is approximated by step waves and the slant component, which shows the smoothness of pictures, is approximated by delta modulation. A new method is proposed here: a statistic adaptation in each block and a constant speed scanning are applied for input picture signals. Finally, it is demonstrated that computer simulations prove the usefulness of the method.