[TriLUG] Spamassassin bayesian filter question

[Douglas Ward]

When I implemented spamassassin with rulesdujour on our e-mail gateway 
server the spam problem dropped so significantly that I've had a hard time 
coming up with a large amount of spam to train the bayesian filter. What a 
terrible problem to have! I now have 200 spam messages that I would like to 
train the message with. I created a mailbox on the gateway server called 
spam and ran the following command:

sa-learn --spam /var/mail/spam

Every message in the inbox file is spam. The answer I get back from 
spamassassin is that it learned from one message (one message examined). 
Wouldn't it search the entire inbox? I moved these files off of our Exchange 
server and to the linux gateway using IMAP and a local pst file. The header 
records were kept intact during the move. Am I missing something?