The invention is related to video compression systems, and in particular to compression/decompression in digital video systems. The present invention discloses a method for compensating the rounding and truncation error when calculating ½ and ¼ position pixels for motion vector prediction. According to the invention, the rounding/truncation offset is attenuated by properly shifting up or down the result of the calculation of the ½ position pixel values. A method alternating occurrences of converting up and converting down interpolated pixel values to nearest integers in the cases where the pixel values are decimals with fractions of 0.5. In a preferred embodiment, each motion vector is assigned either a first or a second notation, so that converting up or down in a predicted block depends on the assignment of the motion vector corresponding to the prediction block. The present invention is particularly useful in the coding standard H.264/AVC.