We were told that the Iraqis don't count. Before the invasion began, the head of US central command, General Thomas Franks, boasted that "we don't do body counts". His claim was repeated by Donald Rumsfeld in November 2003 ("We don't do body counts on other people") and the Pentagon last January ("The only thing we keep track of is casualties for US troops and civilians").

But it's not true. Almost every week the Pentagon claims to have killed 50 or 70 or 100 insurgents in its latest assault on the latest stronghold of the ubiquitous monster Zarqawi. In May the chairman of the joint chiefs of staff said that his soldiers had killed 250 of Zarqawi's "closest lieutenants" (or so 500 of his best friends had told him). But last week, the Pentagon did something new. Buried in its latest security report to Congress is a bar chart labelled "average daily casualties - Iraqi and coalition. 1 Jan 04-16 Sep 05". The claim that it kept no track of Iraqi deaths was false.

The report does not explain what it means by casualty, or if its figures represent all casualties, only insurgents, or, as the foregoing paragraph appears to hint, only civilians killed by insurgents. There is no explanation of how the figures were gathered or compiled. The only accompanying text consists of the words "Source: MNC-I", which means Multi-National Corps - Iraq. We'll just have to trust them.

What the chart shows is that these unexplained casualties have more than doubled since the beginning of the Pentagon's survey. From January to March 2004, 26 units of something or other were happening every day, while in September 2005 the something or other rose to 64. But whatever it is that's been rising, the weird morality of this war dictates that it is reported as good news. Journalists have been multiplying the daily average of mystery units by the number of days, discovering that the figure is lower than previous estimates of Iraqi deaths, and using it to cast doubts on them. As ever, the study in the line of fire is the report published by the Lancet in October last year.

It was a household survey - of 988 homes in 33 randomly selected districts - and it suggested, on the basis of the mortality those households reported before and after the invasion, that the risk of death in Iraq had risen by a factor of 1.5; somewhere between 8,000 and 194,000 extra people had died, with the most probable figure being 98,000. Around half the deaths, if Falluja was included, or 15% if it was not, were caused by violence, and the majority of those by attacks on the part of US forces.

In the US and the UK, the study was either ignored or torn to bits. The media described it as "inflated", "overstated", "politicised" and "out of proportion". Just about every possible misunderstanding and distortion of its statistics was published, of which the most remarkable was the Observer's claim that: "The report's authors admit it drew heavily on the rebel stronghold of Falluja, which has been plagued by fierce fighting. Strip out Falluja, as the study itself acknowledged, and the mortality rate is reduced dramatically." In fact, as they made clear on page one, the authors had stripped out Falluja; their estimate of 98,000 deaths would otherwise have been much higher.

But the attacks in the press succeeded in sinking the study. Now, whenever a newspaper or broadcaster produces an estimate of civilian deaths, the Lancet report is passed over in favour of lesser figures. For the past three months, the editors and subscribers of the website Medialens have been writing to papers and broadcasters to try to find out why. The standard response, exemplified by a letter from the BBC's online news service last week, is that the study's "technique of sampling and extrapolating from samples has been criticised". That's true, and by the same reasoning we could dismiss the fact that 6 million people were killed in the Holocaust, on the grounds that this figure has also been criticised, albeit by skinheads. The issue is not whether the study has been criticised, but whether the criticism is valid.

As Medialens has pointed out, it was the same lead author, using the same techniques, who reported that 1.7 million people had died as a result of conflict in the Democratic Republic of Congo (DRC). That finding has been cited by Tony Blair, Colin Powell and almost every major newspaper on both sides of the Atlantic, and none has challenged either the method or the result. Using the Congo study as justification, the UN security council called for all foreign armies to leave the DRC and doubled the country's UN aid budget.

The other reason the press gives for burying the Lancet study is that it is out of line with competing estimates. Like Jack Straw, wriggling his way around the figures in a written ministerial statement, they compare it to the statistics compiled by the Iraqi health ministry and the website Iraq Body Count.

In December 2003, Associated Press reported that "Iraq's health ministry has ordered a halt to a count of civilians killed during the war". According to the head of the ministry's statistics department, both the puppet government and the Coalition Provisional Authority demanded that it be stopped. As Naomi Klein has shown on these pages, when US soldiers stormed Falluja (a year ago today), their first action was to seize the general hospital and arrest the doctors. The New York Times reported that "the hospital was selected as an early target because the American military believed that it was the source of rumours about heavy casualties". After the coalition had used these novel statistical methods to improve the results, Blair told parliament that "figures from the Iraqi ministry of health, which are a survey from the hospitals there, are in our view the most accurate survey there is".

Iraq Body Count, whose tally has reached 26,000-30,000, measures only civilian deaths which can be unambiguously attributed to the invasion and which have been reported by two independent news agencies. As the compilers point out, "it is likely that many if not most civilian casualties will go unreported by the media ... our own total is certain to be an underestimate of the true position, because of gaps in reporting or recording". Of the seven mortality reports surveyed by the Overseas Development Institute, the estimate in the Lancet's paper was only the third highest. It remains the most thorough study published so far. Extraordinary as its numbers seem, they are the most likely to be true.

And what of the idea that most of the violent deaths in Iraq are caused by coalition troops? Well according to the Houston Chronicle, even Blair's favourite data source, the Iraqi health ministry, reports that twice as many Iraqis - and most of them civilians - are being killed by US and UK forces as by insurgents. When the Pentagon claims that it has just killed 50 or 70 or 100 rebel fighters, we have no means of knowing who those people really were. Everyone it blows to pieces becomes a terrorist. In July Jack Keane, the former vice chief of staff of the US army, claimed that coalition troops had killed or captured more than 50,000 "insurgents" since the start of the rebellion. Perhaps they were all Zarqawi's closest lieutenants.

We can expect the US and UK governments to seek to minimise the extent of their war crimes. But it's time the media stopped collaborating.