P.O. Boxes are of course not tied to a physical location, so there's nothing you can do about those. As far as the other dirty data, some of the other not-free geocoders that I've worked with, mainly Microsoft Mappoint Web Service, can handle a little more garbage (although I think they don't like # signs either). But to use the Yahoo one, you may have to do some string filtering, etc. as you mentioned. Sorry! That's a product of Yahoo's functionality/flexibility.