As　an　efficient　video　feature　extraction　algorithm,　video　fingerprints　are　widely　used　in　video　retrieval　systems.　However,　the　results　derived　by　the　algorithm　are　significantly　impacted　by　noises　in　video　sequences.　Analysis　from　150　video　sequences　indicates　that　the　performance　of　video　fingerprints　is　decreased　in　two　major　situations.　First,　offsets　of　color　and　brightness　are　variable　among　video　frames;　Second,　abrupt　intense　disturbances　occur.　To　tackle　this　problem,　we　proposed　an　improved　algorithm.　The　new　algorithm　compares　the　changes　of　video　fingerprints　instead　of　comparing　video　fingerprints　directly　and　an　exception-factor　is　set　to　deal　with　abrupt　intense　disturbances.　Experiments　demonstrate　the　proposed　algorithm　produces　more　accurate　results　than　original　one.