At a restricted government facility in New Mexico, scientists have set off more than 600 explosions in closed chambers to study dirty bombs.
Slow-motion video shows how one would work — blowing up conventional explosives to spread dangerous radioactive material.
In the worst case, radiation could contaminate dozens of city blocks, taking years to clean up, leaving people wary of returning even longer.
"That's the kind of high consequence attack that would have a real impact on our national economy and our way of life," says Homeland Security Secretary Michael Chertoff.
Fred Harper, a nuclear engineer at Sandia National Labs near Albuquerque, says his tests have produced some good news. Building a dirty bomb turns out to be very difficult.
"It's more complicated than most people would assume," he says. "It depends on the material properties and the device design. And there is a lot of ways to blow it."
Even so, the fear that a terrorist might get it right is a major worry for Homeland Security, which is now cranking out radiation monitors and hoping to set up a ring of detectors around New York and other big cities. Government researchers are also testing more advanced detectors, which are easier to deploy.
Radiation monitors are already working around sensitive sites, such as the White House in Washington and the stock exchanges in New York. But the materials they're designed to detect are widely available.
The U.S. has licensed more than 20,000 users of radioactive materials — for cancer treatment machines, for example, and oil field logging equipment.
Government investigators warn that's a huge shortcoming. Despite the alarm about dirty bombs, experts say, those materials are nowhere near well enough secured.
"These materials can either be stolen or perhaps acquired by staff members that may be associated with the wrong types of people," says NBC terrorism analyst Michael Sheehan.
Security experts call for tighter controls on nuclear materials to prevent the kind of attack that Homeland Security hopes never comes.