Audiotrieve's InBoxer 2 strains out the spam from any POP3, IMAP, or Exchange e-mail account using Bayesian analysis. It works with Outlook 2000 or later (not Outlook Express) and supports all modern Microsoft Windows versions. And its plug-ins offer additional features you won't find in other products, some at no extra charge.

InBoxer's technology is derived from speech recognition, and like speech-recognition software, it needs training to learn precisely what you consider to be spam. It assigns a spam percentage to each message—under 15 percent is a ticket into the in-box, while over 90 percent gets the message blocked. (Advanced users can tweak these cut-off values). Messages with an in-between score go to a Review folder, which you must process manually, clicking on the "Keep" or "Block" button to rescue or trash each message. If the program allows spam into the in-box or blocks a valid message, you absolutely must correct it—otherwise InBoxer will assume the categorization was correct and learn accordingly.

To jump-start the learning process, you can supply InBoxer with folders containing examples of valid mail and spam. If desired, you can add a spam-percentage column to Outlook's display, or get a word-by-word analysis of any message. Adding a message's sender (or entire domain) to the whitelist or blacklist is a snap; you can also import your Outlook Contacts into the whitelist.

If your mail runs 90 percent spam, you've probably already turned off Outlook's new-mail notification. InBoxer's MailCall plug-in is a better solution—it provides audible notification only for good mail only, never for spam. The MailTones plug-in kicks the concept up a notch with the equivalent of custom ringtones for different message types. You might define one sound for a message from the boss and another for a message with "party" in the subject. The PopBoxer and Redirector plug-ins, available at extra cost, keep spam from reaching your BlackBerry or other handheld e-mail device.

Unfortunately, our testing ran afoul of InBoxer's learning mode. It took several days to review the initial influx of 10,000 messages, and doing so actually reduced overall accuracy. On instruction from InBoxer, we reset the learning module and trained the product in a more realistic way, reviewing incoming e-mail several times a day.

When we turned it loose to analyze mail without help, it was quite accurate about the messages it was sure of—it put less than 1 percent of valid mail in the Blocked folder and allowed less than 2 percent of spam into the in-box. But it put 38 percent of the valid mail and 11 percent of the spam into its Review folder. In practice, ongoing training should reduce the number of messages for review; judicious whitelisting would help as well.

If you like a hands-on approach to spam-blocking and want to know why a given message was blocked or allowed, InBoxer 2 may be just the thing for you. And we like the plug-ins that offer customized audio notifications and spam cleaning for your handheld e-mail devices. Still, other products manage to identify spam without requiring quite as much "audience participation."