The initiators of the Iraq Body Count, which has produced the lowest figures, were no advocates of the invasion. John Sloboda, an Oxford-based psychologist who co-founded the group, was provoked to do so precisely because he saw no official institution willing to count the human cost of the war. His organisation decided to use "passive surveillance", a statistically conservative method that only deals with facts on the ground. The IBC lists all cases where at least two media sources report an incident causing one or several deaths, keeping a careful tab of the victims' age, gender, occupation, manner and place of death, where information is available. In addition to providing a running total (which now numbers just under 90,000), its figures have allowed the IBC to produce some of the best data and analysis on how the Iraq war has changed over five years. They reveal that the Americans killed four times more civilians in the first two years of the war (thereby provoking armed resistance to the occupation) than al-Qaida-linked insurgents did, in spite of the media's emphasis on car bombs and suicide attacks. They also show the explosion of criminal violence since 2003, one byproduct of the removal of Saddam Hussein's draconian security methods - or what Bush would call Iraq's new "freedom".

The Americans learned one lesson from Vietnam: don't count the civilian dead. As a result, no one knows how many Iraqis have been killed in the five years since the invasion. Estimates put the toll at between 100,000 and one million, and now a bitter war of numbers is raging. Jonathan Steele and Suzanne Goldenberg report
...
Finally, they point out that more recent data confirm their findings and even suggest a higher figure. The British polling firm Opinion Research Business (ORB) asked 1,720 Iraqi adults last summer if they had lost family members by violence since 2003; 16% had lost one, and 5% two. Using the 2005 census total of 4,050,597 households in Iraq, this suggests 1,220,580 deaths since the invasion. Accounting for a standard margin of error, ORB says, "We believe the range is a minimum of 733,158 to a maximum of 1,446,063."