@jpipes: Thanks a lot ! This is exactly what we are looking for. A first test:

two tables containing 1.2 million rows (temp and ods). After changing 47.000 rows on the temp. I've ran the update statement. (server Pentium III 500 Mhz 384Mb an old one to test on) completed in 27 seconds! Comparisons against the cursor statement will also be made.

@SeekQuel: Very nice (the checksum). Does this neat little "trick" offers an important performance gain ? (anyway, it is shorter and more readable as we have tables with 280 columns)

@jeremy-2 Kemp: In our situation the set based solution seems very fast. Remember: the set only works on the dataset results from the joins

update 1..5% of the data

insert 1..5% of the data

instead of sequentially digging through the whole 100%