In　order　to　improve　the　efficiency　of　ETL　workflow　executing,　this　paper　presents　a　distributed　ETL　engine　based　on　MAS　and　data　partition　technology　and　also　researches　the　methods　of　partitioning　the　massive　data　stream　in　both　horizontal　and　vertical　ways.　The　engine　referred　to　will　partition　an　ETL　workflow　which　meets　the　conditions　of　being　partitioned　into　multiple　sub　workflows　for　parallel　executing.　Each　of　the　sub　workflow　is　executed　by　an　agent,　so　that　multiple　agents　could　work　together　to　complete　the　collaborative　work.　Experimental　results　show　that　this　system　has　good　scalability　and　could　well　improve　the　efficiency　of　ETL　workflow　executing.　©　2011　IEEE.