POP3Filter makes use of an advanced bayesian classifier as its primary method of classifying messages as spam.
Bayesian classifiers in general are recognised as one of the best methods for catching spam as they use statistical methods to learn from the types of both spam and non spam email that you receive. Because of their learning nature, they automatically adapt to the ever changing face of spam, rather than needing regular updates like ordinary rule based spam filters require. Also, because of their personalized nature, spammers are unable to test that their spam will get past bayesian filters before they send it out.
Enhancements to the bayesian classifier found within POP3Filter include automatic learning from incoming messages and the ability to discriminate likely spam words even when spammers try to hide or obscure them - e.g. by spelling "viagra" as "v1@gra"
The bayesian classifier in POP3Filter alone accounts for around 98-99% of all spam messages being caught.