> tried it and settled on bogofilter now. SpamAssassin needed much
> more training than I could provide it to kick in. It would also
> require some tweaking because otherwise it classifies all email
> forvarded form Usenet as forged headers, resulting in useless
> scores. Not to mention SpamAssasin eats more resources than
> mozilla and runs about equally fast.
>
> So if you want just a simple bayesian filter that is quite usable
> with very little training and setup try
> 	http://bogofilter.sourceforge.net


Try also POPfile if you're using POP to get your mail.  Works with an
arbitrary number of buckets, not just "spam"/"not spam", though you can do
that too if you want.