At this point, assuming you have sent mail as prescribed above, and some of those messages were returned, you will have one or more messages in your bounce box.  Each of these messages will be addressed to bounce_<id>@yourdomain.com, where <id> represents the id of the e-mail address in your database which is related to the bounce.

Now it is important to understand that there are two types of bounces: hard and soft.  Permanent failures, such as a nonexistent account or domain, are considered hard bounces.  Other failures, such as a full mailbox or blocked domain, are considered soft bounces.  Instead of flagging your addresses as good or bad, your database can keep a running count of hard and soft bounces for each address.  That way, your mailing application can be more intelligent about determining which addresses to exclude from future mailings.  For example you might only want to send mail to any addresses with less than 8 soft bounces and less than two hard bounces.  I usually do not like to exclude someone from future mailings unless they have more than one hard bounce.  Just to be sure that the address is really invalid, I look for at least two hard bounces. 

Your application will have to scan the text of the bounced messages looking for phrases that indicate the reason for the bounce.  It will look for such phrases as "delivery failure", "box full", etc...  (The downloadable sample code includes a database of the phrases we have discovered in typical bounced messages.)  Your app will determine if each bounce is hard or soft based on the phrase it finds in the message.

Once your app determines if the bounce is hard or soft, it can increment the bounce_hard and bounce_soft fields in the database accordingly.  It can then delete the message from the bounce box.  If your app can not determine if the message is a hard or soft bounce the message can be left in the bounce box.  Periodically the messages remaining in the bounce box can be analyzed by a human who can visually determine why they were not identified by the phrase scanner algorithm.  The algorithm can then be updated to catch this type of message.  Once your app is run again, it should handle this message properly and clear it from the bounce box.  As time goes on, your phrase scanning algorithm should improve more and more. If you start with the phrases included with the downloadable sample code, your app should immediately id just about every bounced messages.