Messages are automatically learned to the system. If you in some way decide to feed messages yourself to improve the score, it's possible to save complete messages (including headers) to the correct folders.

If you want to manually finetune the bayesian data base (which is always good since only human can really detect if a message is spam or not) you can put the messages in the ./ham folder for non spam and in the ./spam folder for spam. Then press the 'Learn now' button on the Filtering tab and the messages are learned to the system.
If the messages 'Learned x from y total' appear, the learning is finished. Remember that x and y do not need to have the same value, since messages can only be learned once, and they might be skipped here if they have been learned automatically before.

NOTE: Messages must ALWAYS be saved to the folders with there complete sourcecode!! So do not copy the message from you emailreader screen since a lot of information is lost then and the learning will make no sense.

Also make sure to not only manually learn spam messages or only ham messages. There should be a certain balance between both groups of messages.
A preview of a small message including headers will be shown below. Obviously this is a spam message and should be copied to the ./spam folder.