A video decoding and video encoding method of performing inter prediction in a bi-directional motion prediction mode, in which a prediction pixel value of a current block may be generated by not only using a pixel value of a first reference block of a first reference picture and a pixel value of a second reference block of a second reference picture, but also using a first gradient value of the first reference block and a second gradient value of the second reference block, in a bi-directional motion prediction mode. Accordingly, encoding and decoding efficiency may be increased since a prediction block similar to an original block may be generated.