Like many other people I have experienced a massive increase of spam in one of my public email accounts over the last year. For that account my junk mail rate is currently at about 250 messages per day. Until recently Mozilla's bayesian filter was quite helpful weeding out these annoying messages. However, I was never able to get a higher spam detection rate than 90%, even after proper training with thousands of messages. Even though 90% is pretty good, it is still not good enough at these high spam rates. After a long weekend and 1000 spam mails later you will still have 100 spam messages waiting in your inbox. That is not acceptable. On top of that I had quite a significant number of false positives.
Looking for alternatives I came across SpamProbe, an advanced bayesian filter for Unix-based systems that Paul Graham mentioned in one of his recent articles. SpamProbe's author, Brian Burton, claims that his software can offer detection rates of 99% and higher. Having used SpamProbe for over two weeks I'm already very close to that detection rate. On top of that I haven't found one single false positive after the switch. As you can imagine I'm very pleased with this great piece of software.