Well, well, well. A nicely written article that makes tons of sense. Recently I have had to write some complex T-SQL conversion scripts with multiple batches and a lot of logic. Getting the bugs worked out of them once written was tough to do and would have been a lot tougher without proper error handling and knowing about the limitations of XACT_ABORT. Eventually I ended up turning it off like the author suggested and using a lot of @@error logic, and yes like Don said you unfortunately have to work out all the kinks during testing because the transactions are aborted when using batches whenever the first error occurs. Can't wait for the try-catch stuff in 2005.