bayes issue

[Jon Fraley <jf...@glenraven.com>]

My current setup: SpamAssassin version 3.0.2 running on Perl version
5.6.1.  Everyday I do an /usr/bin/sa-learn on a ham folder and a spam
folder for which users drop emails in.  However this morning after
running sa-learn against the folders my nham and nspam have both dropped
below 200.  Now of course Bayes is not working.

I saved all the emails that users have sent to these two folders to be
able to rebuild if necessary and have done so in the past.  Now when I
try to learn from the saved messages it says it has learned from 

spam messges:
Learned from 12452 message(s) (13070 message(s) examined).
ham messages:
Learned from 2667 message(s) (3288 message(s) examined).

But nham and nspam is still well under 200 each.  What can I do?

Thanks,

Jon