Abstract: Based　on　138evenl y　distributed 　IGS　coordinate 　time 　series 　spanning　more　than　10years site －to－site 　coefficents 　in 　lar ge　scale　GPS　network　were　anal yzed．Their 　site －to－site 　coefficients 　indicate si gnificant 　corelationshi p　even　when　the　between－site 　distance　more　than　5　000km．Thereforethis paper 　proposes 　a　new　method　introducing　coefficients 　as　weight 　factor 　to 　calculate 　common　mode　errorand 　it 　improved 　the 　correlation－based　spatial 　filterin g　techni que．Time　series 　from　global 　and　regional 　networks　were　used 　to　verif y　this　new　method．The　results 　show　that 　this 　improved　coefficientbased　spatial 　filterin g　techni que　overcomes 　the 　spatial 　scale 　limit 　and　effectivel y　extracts 　common　mode error 　of　GPS　coordinates 　time 　series．