A method, apparatus and system for decoding a bi-directional picture predicted from two reference frames with a motion compensation codec that is capable of locating reference frame portions from only one reference frame at a time. Including locating a first reference frame portion from a first frame in a first pass, storing the first reference frame portion in memory; locating a second reference frame portion from a second frame in a second pass, and combining the first reference frame portion and the second reference frame portion to form a bi-directionally predicted portion. Also, a method and apparatus for restricting the encoding of a bi-directional picture predicted from two reference frames to be decoded by a motion compensation codec that is capable of locating reference frame portions from only one reference frame at a time and is capable of decoding a limited number of partitions for each macroblock.