The company processes more than half a terabyte of traffic data every hour, 80 billion queries a day, or about 2 percent of all the traffic in the world, according to Andrew Hay, director of security research at OpenDNS.
The company also has access to several years' worth of stored traffic data.
"Very few companies can say that they see this kind of traffic," he said.
The algorithm, called SPRank, uses mathematical concepts most often used to analyze sound waves, such as in Pandora's Music Genome Project.
In particular, the company analyzed traffic to known domains to see how traffic patterns changed when the domain became malicious.
There are other security companies that have begun analyzing network activity to spot suspicious patterns, but typically within the constraints of a single enterprise network, rather than the entire Internet.
According to OpenDNS, the new algorithm can identify hundreds of new compromised domains every hour, over a third of which are not detected by any other antivirus or anti-malware vendor.
At that point, the system also starts looking at other nearby and associated domains and decides whether to block just the domain, the IP address, or the entire network.
"As we iterate and pivot on that information, we may find additional domains and IPs that we want to block," Hay added.
This second stage, which the company calls Predictive IP Space Monitoring, can find malicious domains even before they are used for actual attacks.
OpenDNS has been slowly adding in the functionality for its existing customers, and officially launched the system this morning.
"We've been very safe and taking a very casual stance as to what we're putting in there, we want to be sure that the false positive rate is very low," he said.
The new pattern-spotting system is only one of several methods that OpenDNS uses to spot malicious domains. Other innovative approaches the company has rolled out include using natural language processing to spot keywords that frequently appear in phishing domain names, and reverse engineering domain generating algorithms used by malware authors.
OpenDNS also looks for the use of uncommon domain name registrars, where the domains are hosted, where most of the visitors are located, and Whois information that doesn't match that of the parent domain.
"We keep looking at our traffic and looking for new ways to look at it," said Hay.