A method and device for predicting a block of pixels of an image for encoding/decoding same. The method comprises a step (1) of defining a causal vicinity of said block to be predicted, a step (2) of searching for candidate patches during which a set of candidate patches is formed from at least one patch belonging to an image other than the image to which the block to be predicted belongs, each patch being formed from a block and from a causal vicinity of said block, and a step (3) of block prediction during which the block is predicted at least from the block of at least one patch of said set of candidate patches.