Disclosed herein is a method of decoding an image. The method includes determining a prediction mode of a current block as a merge mode with motion vector difference (MMVD), deriving a merge candidate list of the current block, deriving a motion vector predictor of the current block using the merge candidate list, deriving a motion vector difference of the current block, and deriving a motion vector of the current block using the motion vector predictor and the motion vector difference. Only some of candidates in the merge candidate list are used to derive the motion vector predictor.