A prediction signal is efficiently generated with a small amount of additional information and a low degree of computational complexity. An image prediction encoding device (100) includes a prediction signal generator (103) for generating a prediction signal corresponding to the object pixel signal of one of the object regions defined by division by a block divider (102), a motion estimator (113)for searching for a motion vector required to generate the prediction signal from a reproduced signal, and a subtracter (105) for generating a residual signal indicating the difference between the prediction signal and the object pixel signal. The prediction signal generator (103) includes a candidate prediction signal combining unit (203) which determines a search region which is a partial section of the reproduced signal according to the motion vector, generates one or more type-II candidate prediction signals from the search region, and generates a prediction signal by processing the type-II candidate prediction signals and a type-I candidate prediction signal generated according to the motion vector by a predetermined combining method.