Molecular　docking　technology　is　an　effective　approach　for　prediction　of　intermolecular　interactions　and　recognition.　The　design　of　a　scoring　function　for　selecting　near-native　structures　is　very　important　for　successful　prediction　of　complex　structures.　In　this　article,　the　main　computational　methods　for　scoring　items　in　protein-protein　docking,　such　as　geometric　complementarity,　contact　area,　van　der　Waals＇　interaction,　electrostatic　interaction,　and　statistical　pair　propensity　potential,　are　reviewed.　Including　our　work,　we　introduce　commonly　used　scoring　schemes　and　some　strategies　in　screening　decoys　based　on　the　information　for　protein　binding　sites.　The　characteristic　scoring　functions　in　the　commonly　used　docking　programs　are　compared　and　summarized.　The　major　problems　in　the　existing　scoring　function　in　protein-protein　docking　are　discussed　along　with　prospect　for　future　research.