Bayesian filters analyze the content of emails by examining the words and phrases used. Each word is given a probability score of being found in spam versus legitimate emails. These scores are then combined to give an overall probability that an email is spam. The algorithm learns and improves over time as it processes more emails.