Bayesian filtering is based on tokens. It uses those tokens to recognize spam or not spam and allows each Barracuda to be customized to a particular networks needs. For example, perhaps I get a lot of aviation email. So I mark hundreds of emails as not spam and the bayesian filtering will start to see a pattern of words like fixed wing and helicopters. But say your network doesn't care about aviation. You mark spam messages that come through with those words as bad and eventually the bayes filtering knows that if the email has fixed wing or helicopters in it, it is most likely spam.Here is a really good article on bayesian filtering: http://email.about.com/cs/bayesianfilters/a/bayesian_filter.htmMake sure your scoring is set such that it will block the messages. The bayes filters simply determine odds of something being spam and score it accordingly.