Bayesian filtering is a statistical approach to spam filtering. It allows the system to calculate a spam probability for every email you receive. This probability is calculated on past email you have received as well as past spam and is calculated on a user-by-user basis so only your past email is used to predict whether an incoming message is spam. After all, what someone else might consider legitimate you might consider spam. You may click here for a more complete explanation of Bayesian filtering.

No. Your email is constantly purged from our system within 1-2 days of you downloading your mail from us. When Bayesian statistics is enabled, the system will maintain a statistical file that counts how many times each word has appeared in your real email and in spam. That is all Bayesian needs to know to work. There is no way to determine the contents of email messages once they have been purged. All the Bayesian system will keep is statistical information that indicates how many good and bad messages contained each word.

No. Bayesian filtering is very easy to use. The only thing you have to do is report any spam that gets through the filter and also report any email that is incorrectly deemed to be spam. As long as you report these incidents your Bayesian filter will improve in accuracy and you'll have fewer reports to make.

According to Paul Graham, Bayesian filtering can achieve a 99.5% success rate. This means only 5 of every 1000 spams get through. If you receive 20 spams per day, Bayesian filtering should be able to reduce this to one spam every 10 days or so.

We suggest that you enable as many filters as possible when you first start using Bayesian filtering. Because your Bayesian filter must first accumulate a history of good email and spam, Bayesian filtering will not activate the moment you select it. Rather, the filter will start accumulating statistics immediately and will implement Bayesian filtering when it has sufficient statistics to do so. You will receive an email to notify you when Bayesian filtering starts operating. Until then, enabling the other filters will allow the system to catch most of your spam which itself will help teach your Bayesian filter. Once the Bayesian filter is operating you may still want to leave the other filters enabled since we may be able to help you filter out known spam websites that may not be part of your Bayesian statistics.

While no filtering method is 100% perfect, Bayesian filtering should be the most accurate at catching spam while letting real email through. When you first start using Bayesian filtering it is possible that some spam will get through and that, possibly, some valid email will be blocked. This is why you must report it to the automated system when it happens--so the Bayesian filter can adjust accordingly. As you continue to use the Bayesian filter, however, more and more spam will be caught and fewer and fewer real emails will be blocked. Of course, you may wish to establish whitelists in the event you see certain messages from certain people being blocked. This will also help the Bayesian filter learn that emails from these sources are not spam.

Remember that when you first enable the Bayesian filter all that will happen is that the filter will start to collect statistics. The Bayesian filter can't operate until it has sufficient statistical information on which to base its filtering decisions. By default, the Bayesian filter will start filtering when it has at least 100 good messages and 100 spam messages in its statistics file. Thereafter, the Bayesian filter will become more and more accurate as more and more statistics are generated. The Bayesian filter will start out moderately accurate and improve from there. With statistics for 1000 good messsages and 4000 spam messages we have seen the Bayesian filter catch 99.8% of the spam received. So give the Bayesian filter time--if you report missed spam as well as false positives you will see accuracy improve drastically.