The company that the Florida secretary of state contracted with in 1998 to help purge the state rolls of ineligible voters is well connected to GOP circles. The chairman of the board of Database Technologies, now the DBT Online unit of ChoicePoint Inc. of Atlanta, was former astronaut and prominent Republican Frank Borman.

Also on the board were billionaire Ken Langone, who was co-chairman of the fundraising committee for New York City Mayor Rudolph Giuliani's aborted US Senate race; and big GOP funder Bernie Marcus, co-founder of Home Depot. Howard Safir, former New York City police commissioner under Giuliani, is a consultant to the company, and Vin Weber, best known as Newt Gingrich's legislative enforcer when the pair controlled the House of Representatives, is the company's lobbyist. The company says that it favors no party.

A report that Secretary of State Katherine Harris had ordered the removal from voting lists of 8,000 Florida citizens, every one of them wrongly identified by DBT as felons from Texas, first appeared in Britain's Observer, where I work as an investigative reporter. Harris and ChoicePoint claimed at that time to have corrected their erroneous ways, but the Observer, with the help of a team from Salon, reported in December that, extrapolating from known figures, at least 15 percent of the 58,000 felons named on the new scrub lists had also been wrongly identified as felons.

A ChoicePoint spokesman termed the British reports a "vile, lying, inaccurate pack of nonsense." A more upbeat spokesman noted with pride that "fifteen percent wrong [is] eighty-five percent right!" In a later statement the company said the scrub list contained the names of potential felons only, and that "Florida law prevents names from being removed from the voting roll unless the information is confirmed by local officials–not by us."

One county (Leon, which includes Tallahassee) allocated resources to verify independently the criminal records of those on the list who lived in that county. According to officials there, it could confirm only thirty-four of 694 names, indicating that the error rate could be as high as 95 percent, not 15 percent.