The use of motion vectors to improve the efficiency of the prediction of pixel values. The prediction uses motion vectors to provide offsets into past and/or future reference frames containing previously decoded pixels that are used to form the prediction and the error difference signal.