> Is there anything I'm missing that is preventing it from using the index? It > just seems weird to me that other joins like this work fine and fast > with indexes, > but this one won't. Did You consider clustering both tables on the dsiacctno index? I just checked that for a 4M rows table even with enable_seqscan=on and default *page_cost on PG 8.1.4 an index scan is being chosen for select * from table order by serial_pkey_field This is essentially the question in Your case - sort it, or get it sorted via the index at the expense of more random IO. I think clustering should work for You, but I am no expert, check with others. Greetings Marcin
