An image decoding method comprising the steps of: receiving an encoded bitstream including information of I and P frames; and executing motion compensation by synthesizing a predicted image of a current frame using motion vector information included in the encoded bitstream and a reference image which is a previously decoded image; wherein said motion compensation includes calculating intensity values at points where no pixels actually exist in the reference image by bilinear interpolation, the bilinear interpolation being performable using a positive rounding method and a negative rounding method, wherein the bilinear interpolation is performed using the rounding method specified by rounding method information included in the encoded bitstream of the current frame when the current frame is a P frame, and wherein the rounding method information is not received from the encoded bitstream of the current frame when the current frame is an I frame.