New research into the viral spread of email chain letters suggests that the messages travel in a less direct and more diffuse pattern than previously assumed.

Jon Kleinberg, of Cornell University, and David Liben-Nowell, of Carleton College, found that the messages did not spread like a virus with each producing many direct "descendents" in a tree pattern.

Instead, people seem to be selective in forwarding messages to others in their social networks.

For example, the researchers discovered that the messages produced only a single descendent 90 per cent of the time.

The research, supported by the National Science Foundation, Google, Yahoo and the MacArthur Foundation, examined the way in which two email petitions circulated over a period of 10 years.

One petition in support of public radio began circulating in 1995, and the other, in opposition to the US invasion of Iraq, originated in 2002.

The messages had the common characteristic of being widely circulated; the researchers found 316 copies of the public radio petition with more than 13,000 signatures, and 637 copies of the Iraq petition with almost 20,000 signatures.

Using this data, the researchers mapped out how these messages travelled from recipient to recipient on a tree pattern.

A careful analysis of the pattern challenges some of the common assumptions about how messages spread, including the viral contagion theory.

These messages also rarely took the most direct route between two inboxes, even when two people were connected by a few degrees of separation.

"The chain letters themselves often got to people by highly circuitous routes," said Kleinberg.

"You could be six steps away from someone, yet the chain letter could pass through up to 100 intermediaries before showing up in your inbox."