A video decoding method, according to the present invention, comprises the steps of: deriving an L0 motion vector and an L1 motion vector of a current block; deriving an L0 prediction sample and an L1 prediction sample for a first position inside the current block on the basis of the L0 motion vector and the L1 motion vector; determining whether to apply a bidirectional optical flow to the current block; deriving a refined motion vector for a subblock which comprises the first position, when it is determined to apply the bidirectional optical flow; deriving a refined prediction sample for the first position on the basis of the refined motion vector; and obtaining a prediction sample for the first position by using the L0 prediction sample, the L1 prediction sample, and the refined prediction sample.