Very odd indeed! I don't see CLR RI anywhere in your original post. Well, we'll pretend it's really there. The issue still remains, which is that SBUF will not contain any new data from which to "refresh" until RI becomes set again. It's all very apparent when you read how the serial ports work and look at their block diagrams. I guess there's nothing to prevent you from sitting there looping in your ISR polling SBUF waiting for some data change and ignoring RI -- I've just certainly never seen anybody do anything like it before. But once you get your "refreshed" data after polling for it, RI will have become set again and you must clear it again to allow any further serial reception.