Neil, your question has (at least) two aspects: 1. SIZE AUTO and which columns get histograms applied Oracle only considers columns used by single-table access predicates for automatic generation of histograms- columns only used as part of join predicate as far as I know will not be considered as histogram candidates 2. SIZE AUTO and the generated histogram bucket size There are cases where the automatically generated histogram bucket size might be sub-optimal, which means that a column has less than 254 distinct values, but still a height-balanced histograms gets generated instead of a superior frequency histogram. You can notice this if the bucket size is smaller than the number of distinct values of the column. A frequency histogram might help here to get better estimates. In general, histograms on join columns not always change the outcome for the better, even in case of skewed column data. In particular with height-balanced histograms the results can easily get worse. For an in-depth discussion, see Alberto Dell'Era's paper: http://www.adellera.it/investigations/join_over_histograms/index.html As others have mentioned - can you confirm that the statement performs better with your suggestions? Run the statement with STATISTICS_LEVEL = ALL and DBMS_XPLAN.DISPLAY_CURSOR(...'ALLSTATS LAST') option to compare actual and estimated cardinalities. Regards, Randolf Oracle related stuff blog: http://oracle-randolf.blogspot.com/ Co-author of the upcoming "OakTable Expert Oracle Practices" book: http://www.apress.com/book/view/1430226684 http://www.amazon.de/Expert-Oracle-Practices-Database-Administration/dp/1430226684 ______________________________________________________ GRATIS für alle WEB.DE-Nutzer: Die maxdome Movie-FLAT! Jetzt freischalten unter http://movieflat.web.de -- //www.freelists.org/webpage/oracle-l