The “layered representation” is a method to represent a moving image by relatively moving layered maps in each divided region after they have been superimposed in image depth order, like cell images in an animated picture. The conventional layered‐representation coding method can efficiently represent an object with small movements, such as a background, but the method has difficulty in representing a nonrigid object which changes its texture and shape, like an image of a person. This paper describes a layered‐representation coding method which overcomes this difficulty. The proposed method divides each region along the time axis, and a moving image including nonrigid objects is efficiently reconstructed by mixing maps of frames. This paper also describes an expanded setting representation which can treat three‐dimensional data efficiently. The proposed method can represent a smooth moving image without discontinuous frame shifting, since its image quality can be controlled spatially and temporally. The proposed method is compared with a conventional method (waveform‐similarity method H.263) of a low‐bit rate (less than 32 kbit/s), the former being superior to the latter in a visual evaluation. © 1999 Scripta Technica, Syst Comp Jpn, 30(1): 70–80, 1999