sa-learn with spam

[Hou, Ming]

Hi, 

I am wondering when I use "sa-learn --rebuild --spam a_spam_message", then 
"sa-learn --rebuild". sa-learn printed "Learned from 1 message...", and I 
assumed that SpamAssassin 2.63 will learn this spam by saving to Bayesian 
database. Then, I sent "a_spam_message" through SpamAssassin again, 
"a_spam_messages" was not marked as spam, why?

Thanks,
ming