A Bayesian filter is a program that uses Bayesian logic (also called Bayesian analysis) to evaluate the header and content of an incoming e-mail message and determine the probability that it constitutes spam. Bayesian logic is an extension of the work of the 18th-century English mathematician Thomas Bayes.
Bayesian filters aren't perfect, but because spam characteristically contains certain types of text, such a program can be amazingly effective when it is fine-tuned over a period of time. A Bayesian filter works by categorizing e-mail into groups such as "trusted" and "suspect," based on a probability number (ranging from 0 or 0% to 1 or 100%). The categories are defined according to user preference.
Spammers are constantly trying to invent new ways to defeat spam filters. Certain words, commonly identified as characteristic of spam, can be altered by the insertion of symbols (such as periods), or by the use of nonstandard but readable characters. But as the user instructs a Bayesian filter to quarantine or delete certain messages, the filter incorporates this data into its future actions. Thus a Bayesian filter improves with time, so it becomes more likely to block spam without also blocking desired messages.
Bayesian filters are best used in conjunction with anti-virus programs. Malicious viruses or worms can occasionally appear as attachments to e-mail messages, even from trusted sources.