If you use Outlook, someone wrote a new Bayesian filter for it that seems to work pretty well. I saw it on slashdot the other day and turned a couple of people on to it with decent results.:
http://starship.python.net/crew/mhammond/spambayes/

If you want to know what Bayesian filtering is all about, you can check this out: http://www.smh.com.au/articles/2003/06/23/1056220528960.html