I'm a couple of levels more paranoid than that. First, I'll write the DELETE as a regular SELECT (to preview number of rows), then turn it into a SELECT INTO to

I'm a couple of levels more paranoid than that. First, I'll write the DELETE as a regular SELECT (to preview number of rows), then turn it into a SELECT INTO to save the soon-to-be-deleted rows into a table with a backup_date_ prefix (So old backups can be deleted occasionally). Next, before changing anything, I wrap the statement in a BEGIN TRAN, and ROLLBACK TRAN. After all that, I will finally modify the SELECT INTO into a DELETE statement, run it once while wrapped in the transaction (to verify that the number of rows modified hasn't changed), and then finally run without the transaction to delete the rows. Overkill?