Bayes autolearn non-spam, but not spam

Matt Kettler mkettler at EVI-INC.COM
Thu Aug 11 22:42:44 IST 2005

    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "US-ASCII" character set.  ]
    [ Some characters may be displayed incorrectly. ]

Danny Beland wrote:
> For some reason Bayes autolearn non-spam but not spam. I use MailScanner
> 4.40.11 with SpamAssassin 3.0.2 and sendmail.

You're vulnerable to a mime parser DoS. Upgrade to 3.0.4 to fix that.

 The same setup works fine on
> another server.
> I deleted the bayes files and it still doesn't autolearn spam. It seems
> that I'm missing something...
> I have the following in my spam.assassin.prefs.conf for bayes.
> bayes_path /var/spool/spamassassin/bayes
> bayes_file_mode            0644

I'd suggest fixing that file mode. Use 755 not 644. There are rare instances it
gets used for directory creation. SA won't create regular files with the X bit,
but directories need it.

> use_bayes 1
> bayes_auto_learn 1
> bayes_auto_learn_threshold_spam 4.5
> bayes_auto_learn_threshold_nonspam 0.01

It's pointless to set a spam threshold under 6.0, as SA will in effect ignore you.

SA has a hard-coded requirement that there be at least 3.0 of header hits AND
3.0 of body hits before it will learn as spam. That means that the absolute
minimum threshold is 6.0. You'd have to hack the code to change it.

If you don't have network checks enabled, spam autolearning can be somewhat rare
due to low header rule hits. You might consider adding some header-based rules
from or other sites.

SA also won't autolearn a message as spam if it would have otherwise hit
BAYES_00. This is to avoid self-contradictions. Make sure your training is in
good shape.

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at with the words:
'leave mailscanner' in the body of the email.
Before posting, read the Wiki ( and
the archives (

Support MailScanner development - buy the book off the website!

More information about the MailScanner mailing list