Baysean Analysis Included with versions 2.50 onward. Uses "tokens" within email headers and body to automatically learn features of spam and not-spam messages. Specific to your own email. Looks more complicated / scary than it is. By default, it automatically learns based on output from spamassassin After it has gathered enough data, it will start affecting the scores (eg., adding 2.9 points to a message if it thinks it's 90% likely to be spam) you can fine-tune it with additional bodies of "spam" and "not-spam" or just leave it alone.