Ciao Alberto > I cannot understand row source 8 (FTS) in your plan, why > it should FTS ... unless (wild shot in the dark) it keeps > in "memory" the rowids only while doing (b)+(c) and then > FTS to retrieve the remaining columns. Just an observation... I don't think it is the case because if you have an additional column you will see something like: -------------------------------------------------------------- | Operation | Name | -------------------------------------------------------------- | CONNECT BY WITH FILTERING | | | TABLE ACCESS BY INDEX ROWID | DCH_WORK_SURR_MGMT_ORDERED | | INDEX RANGE SCAN | TEST_INDEX | | NESTED LOOPS | | | BUFFER SORT | | | CONNECT BY PUMP | | | TABLE ACCESS BY INDEX ROWID| DCH_WORK_SURR_MGMT_ORDERED | | INDEX RANGE SCAN | TEST_INDEX2 | | TABLE ACCESS FULL | DCH_WORK_SURR_MGMT_ORDERED | -------------------------------------------------------------- In other words the index-only scan is replaced by a normal table access via rowid. Buona giornata, Chris -- http://www.freelists.org/webpage/oracle-l