Provided is a method and apparatus for encoding and decoding a multi-view image. By predicting a current picture based on at least one of an intra (I) picture of a previous image processing block and an I picture of a current image processing block, which is generated for a different view than a view at which the I picture of the previous image processing block is generated, and encoding the current picture based on the prediction, it is possible to prevent display quality degradation that may occur in images at views other than a base view.