A large size of storage is required for the data obtained in the fields such as image diagnosis, physical measuring of phenomena and so on. In order to deduct this heavy burden, a reversible data compression is always employed for a perfect reconstruction. In this paper, a multi-valued image is transformedto the bit planes whose number is neccesary to exactly express the pixel value of original image. Above transformation process is carried out so as to increase the length of l's run on the bit plane. Based on the information whether the value at the position near the target bit to be compressed is 1or 0, conditional probability on target bit is adaptively changed. Arithmetic coding is carried out by utilizing above formulation. Simulation studies for Girl image(SIDBA) and X-ray image of crotch joint shows the effectiveness of proposed procedure.