Abaca, a startup that launched at last week's Interop NY show, claims to have developed a new approach to spam filtering that guarantees a minimum of 99 percent accuracy.The company backs its claim with a guarantee to credit customers for each day the appliance fails to hit the mark (one false positive out every 20,000 messages).

Instead of focusing on mail content or the sender's reputation, Abaca calculates the probability that an incoming message is spam by looking at the distribution of spam and ham addressed to each recipient.

Recipients that have a large volume of spam addressed to them have a higher probability that a new message is also spam, and vice versa for users with a lower volume of spam messages.

Abaca calls this probability a receiver's reputation, and claims it's more effective than typical header and content analysis or sender reputation techniques. That's because spammers can (and do) manipulate content elements and message origins.

Rather than get into never-ending games of one-upsmanship with manipulative spammers, Abaca tracks the reputation of each user in your organization. Each appliance also reports reputation data back to a central cloud for collective analysis. By combining the reputations of all the recipients of a specific message, Abaca says it can determine the spaminess of that message with a very high degree of accuracy.

I'm not 100 percent convinced of the validity of this approach, in part because of the chicken-or-egg problem. Abaca has to figure out which messages are spam and which aren't before it can assign a reputation to a recipient, which means the company has to do some analysis-and thus falls prey to the same weaknesses of other systems.

Abaca says it does some content analysis, but the company doesn't really say what it is. In an e-mail message seeking clarification, they said in addition to the receiver's reputations "…we also look at where the mail is coming from, if it is forwarded, if it is going to a legitimate mailbox, if we can find the 'real IP' of the sender etc." They don't maintain a database of spam signatures, perform Bayesian filtering or conduct heuristic analysis.

With those reservations in mind, it's certainly true the anti-spam fight can benefit from fresh ideas. If the concept of receiver reputation works as advertised, it would put a powerful tool in the hands of beleaguered IT professionals.

While Abaca is battling spammers, it also has its sights clearly set on Barracuda, which has done an admirable job of penetrating the SMB market with low-cost appliances. Abaca is selling appliances for up to 100 users for $1,495, and up to 1,000 users for $3,495.

The company also has long-term plans to offer a spam filtering service, a la Postini (acquired by Google), Frontbridge (acquired by Microsoft) and MessageLabs.

Abaca is privately funded.