To train the Bayesian analyzer, you should have good e-mail, and spam e-mail. You should have at least 100 of each. If you do not have 100 of each, we recommend that you wait until you do.  
If you do not have any good mail, then make sure you have turned on 'Score and store non-spam messages'  
Choose only the accounts for which you have categorized the e-mail as spam and good. The spam messages should have a red dot next to them, and the good messages should have a green dot next to them. Spam Sleuth will do most of the work automatically, but you need to correct any mistakes it may have made before training.  
Begin Training starts the training. If you have trained on some e-mails already, then any new e-mails will be added to the dictionary.  
Reset Training will erase all the training. We recommend you do this if you have bad training.  
The % Trained will always show 0% when you enter. The % trained will change as it trains.