A method for encoding/decoding an image and a device therefor are disclosed. Specifically, the method for decoding the image by the decoding device, the method includes deriving an intra prediction mode of a current block, constructing reference samples with samples adjacent to an top of the current block and samples adjacent to a left of the current block, and generating a prediction sample, when the intra prediction mode is a vertical/horizontal strong mode, using a reference sample having the same x-coordinate as the sample, a reference sample having the same y-coordinate as the sample, a reference sample adjacent to a bottom-left of the current block, and a reference sample adjacent to an top-right of the current block among the reference samples for each sample in the current block, wherein a weighting value w1 applied to the reference sample having the same y-coordinate as the sample and the reference sample adjacent to the top-right of the current block and a weighting value w2 applied to the reference sample having the same x-coordinate as the sample and the reference sample adjacent to the bottom-left of the current block are set to be different from each other.