Thorkild Sørensen of Copenhagen University Hospital and his colleagues looked at data from the Finnish Twin Cohort Study, in which volunteers filled in questionnaires about their health and lifestyle, first in 1975 and again in 1981. These included questions about height, weight and motivation to lose weight. Even after controlling for smoking and excluding anyone with a chronic illness that could have led to weight loss, Sørensen found that overweight or obese people who intended to lose weight in 1975 and succeeded were nearly twice as likely to have died by 1999 compared with those who had no intention to lose weight and stayed about the same (Public Library of Science Medicine, vol 2, e171).

"I have a great heart and great legs but the rest of me is hopeless," says Stanton Glantz cheerfully. The 59-year-old professor of medicine from the University of California, San Francisco, is just a few supersize colas away from being officially obese, but he reckons he's in tip-top condition. "I just got back from a 350-mile bike trip all over the south-west of the United States," he says. That must have burnt off a few pounds? He laughs. "I lost no weight whatsoever. I was eating like a pig I was so hungry all the time."

The received medical wisdom is that Glantz's weight puts him at risk of a slew of illnesses, from osteoarthritis and cancer to diabetes, hypertension and heart disease. But he is confident that as long as he keeps his weight stable with exercise and eats a sensible diet, the risks are not nearly as serious as they're cracked up to be. Many doctors would disagree, but the balance of scientific evidence may be tipping in Glantz's favour.

Nobody doubts that people living in developed countries are fatter on average than previous generations. But the widely repeated claim that this epidemic of flab is endangering the lives of millions of people is now mired in controversy. Some scientists claim that the most apocalyptic estimates - which put obesity on a par with smoking as a public health hazard - are grossly inflated. They also argue that most of the people who fall into the "overweight" category between normal and obese are perfectly healthy. Stranger still, there is evidence that advising these people to diet may actually put their health at greater risk than being overweight does. Is it time to rewrite the rule book on what constitutes a healthy weight?

The standard way of measuring whether someone is too heavy is to determine their body mass index or BMI (see Graphic). Calculated by dividing your weight by the square of your height, below 18.5 is defined as underweight, 18.5 to 24.9 is normal, 25 to 29.9 is overweight and 30-plus is obese. According to this yardstick, more than half of all adults in the UK are overweight and one-fifth are obese. In the US, around two-thirds of the population weigh too much and a third are obese. Worldwide, there are about a billion overweight or obese people out of a global population of 6.45 billion.

Such figures make shocking reading, but until recently it has been difficult to quantify the effect on public health. Sure, being overweight or obese has been linked with all sorts of health problems. But just how dangerous is it to be too fat?

In March last year a team at the Centers for Disease Control and Prevention (CDC) in Atlanta, Georgia, produced a sobering answer. It reported that overweight and obesity was causing 400,000 premature deaths a year in the US alone (Journal of the American Medical Association, vol 291, p 1238), putting it second only to smoking (438,000) as a preventable cause of death. The CDC later admitted a miscalculation and revised the estimate down to 325,000, but the figure still dwarfs the 85,000 for the third most common cause of premature death, alcohol consumption.

Even as the CDC was working on these figures, however, a few experts were starting to question the consensus about the impact of obesity on public health. In May 2004 Paul Campos, a law professor at the University of Colorado in Boulder, published a book called The Obesity Myth in which he argued that there was no good evidence that being overweight or obese per se is a health risk (New Scientist, 1 May 2004, p 20).

Shortly afterwards a leading obesity researcher told The New York Times that the apparent explosion in obesity in the US was not what it seemed. Jeffrey Friedman of Rockefeller University in New York, who in 1994 discovered the gene for the appetite-suppressing hormone leptin, told the paper the apparent surge in obesity was caused by a large number of already overweight people edging over the arbitrary dividing line into obesity. This produced a headline figure of a 30 per cent increase in obesity, he said, but in reality all that was happening was that the already fat were getting a little bit fatter.

Then, in April this year, another group at the CDC published an alternative estimate of the number of premature deaths caused by being overweight or obese in the US. Using newer data, a team led by epidemiologist Katherine Flegal put the number of excess deaths caused by obesity at a relatively modest 112,000.

That in itself didn't change anything - the previous estimate, after all, was for obesity and overweight combined. But when Flegal's team looked at the overweight category, they found something astonishing. Being overweight correlated with a reduced risk of premature death - there were 86,094 fewer deaths a year in this category than in the normal range (see Graph). In other words, overweight and obesity combined were causing nowhere near 325,000 deaths a year. The net figure was closer to 26,000 (JAMA, vol 293, p 1861).

The research unleashed a storm. One researcher said the quality of the Flegal paper was "pitiful", while another praised its statistical rigour and told New Scientist that the original CDC paper estimating 400,000 deaths per year was "a piece of trash". Conspiracy theorists on both sides lobbed accusations that the other was in the pay of an interest group such as the food or diet industry.

In response to the controversy, the Harvard School of Public Health quickly convened a symposium on obesity and mortality. It was a triumph for the orthodox view. Speaker after speaker seized upon Flegal's counter-intuitive findings about the overweight category as a sign the results were not to be trusted. Most critics argued that the apparent protective effect of being overweight was an artifact, caused by the normal weight category being full of smokers and people with undiagnosed wasting illnesses.

Smoking certainly promotes weight loss and is, of course, associated with numerous fatal conditions. And some serious diseases such as emphysema cause wasting and may go undiagnosed for years. So it seems reasonable to assume that smokers and the undiagnosed terminally ill could be over-represented in the normal weight category. "A lot of these people have diseases that are going to kill them," says Walter Willett, the Harvard professor of medicine who chaired the symposium.

Flegal, however, insists her team went to great lengths to eliminate biases due to smoking or illness. For example, they excluded deaths that occurred in the first three years after subjects were weighed and measured, which should account for much of the effect of undiagnosed illness. Glantz believes epidemiologists should take the findings seriously. "I teach statistics and that paper is a really fine piece of work. It's bomb-proof," he says. "I've read the Harvard criticisms and they just don't make any sense to me."

Flegal was as surprised as anyone by her findings, but when she did some digging she discovered several other studies of BMI and mortality that showed the same pattern, with people in the overweight category less likely to die than either the obese or people with a "normal" weight. One recent analysis, for example, pooled data from 26 separate studies and found that even after controlling for smoking, being overweight was linked to a small drop in the risk of death compared with normal weight (Annals of Epidemiology, vol 15, p 87). "Although people think there's all this evidence out there showing a high mortality risk associated with overweight, in fact the literature doesn't show it," says Flegal.

"If correct, all these worries about a huge fraction of the population being overweight just go out the window," says Glantz. "It's not a trivial problem, but the focus should now be on the severely overweight." He believes the evidence from Flegal and others now demands a change in the healthy BMI range. "The current definition of overweight is not like the speed of light or pi. What was considered as the normal, desirable weight is too low."

But even if Flegal's results stand up, not everyone would accept that the BMI guidelines need to change. They point out that the sole purpose of Flegal's paper was to calculate excess mortality, and that this does not tell the whole story.

One problem with mortality figures is that they do not take people's quality of life into account. "If the only thing you're looking at is mortality you're missing the big picture," says S. Jay Olshansky, professor of epidemiology at the University of Illinois in Chicago. "Overweight people may not be dying so rapidly but they still face serious health problems." There is now good evidence, for example, that medical advances such as drugs to combat hypertension and high cholesterol are keeping overweight people alive in ever-increasing numbers (JAMA, vol 293, p 1868). But that does not mean they are leading healthy, happy lives. "To imply that it's OK to be overweight because we're better at treating its complications is ridiculous," says Olshansky. "It's like saying it's OK to get HIV because we can treat it better."

Another problem with mortality figures is that they may hide the dangers that today's glut of obese and overweight youngsters face as they grow older. David Ludwig is director of an obesity programme at the Children's Hospital Boston, where he has witnessed the dangers at first hand. He says that overweight and obese kids are growing under "extraordinarily unnatural" circumstances. "The possible effects on the long-term health of this generation of children are really frightening," he says. Earlier this year, Ludwig, Olshansky and colleagues forecast that obesity and overweight would reduce average life expectancy in the US by as much as five years during this century (The New England Journal of Medicine, vol 352, p 1138).

Improved medical care may be saving the lives of thousands of overweight people, but that doesn't explain why this group seems to be at lower risk of dying early than people in the normal weight category. Willett points out that there are only a handful of common conditions for which being moderately overweight is known to make you less susceptible - bone fractures in the elderly and breast cancer before the menopause. "Apart from these there's really no common disease where being overweight is beneficial, but a long list of diseases where it has an adverse effect on health," he says. Willett claims there is no plausible biological mechanism that could make being overweight a benefit. As far as he's concerned, this ought to be the final nail in the coffin for Flegal's paper.

Flegal, however, does have an explanation. Too much fat may be a bad thing, but the weight of a human body is more than just fat. There is also "lean mass" - muscle, bone, organs and connective tissue. What Flegal may be seeing is a health bonus from extra lean tissue, rather than extra fat. "I wonder whether what we're picking up is an effect of lean mass, but it's speculation at this stage," she says.

Others claim that all the study does is illustrate the shortcomings of BMI. Judged solely on BMI, Brad Pitt and the fitness enthusiast George W. Bush are both overweight. "It's a crude measure that has very well known limitations," says Howard Strickler, an epidemiologist at the Albert Einstein College of Medicine in New York. "Muscle weighs more than fat, so a healthy athlete will often have a high BMI even though they are actually very fit." He believes the decreased mortality seen in the overweight category may reflect the health benefits of being fit and muscular.

But even that does not provide a complete answer. Surely there are just as many fit people in the normal weight category, if not more. How come overall mortality is slightly higher in this group than in the overweight one? The answer may not be lower levels of lean mass per se, but a surprising effect on people's health when they lose it.

In June, a study by researchers in Denmark and Finland suggested that people who succeed in losing weight may increase their risk of dying prematurely. It is well known that when people lose weight by eating less, they shed lean tissue as well as fat. And there may be a cost from losing lean mass that offsets the benefits of shedding the blubber.

Thorkild Sørensen of Copenhagen University Hospital and his colleagues looked at data from the Finnish Twin Cohort Study, in which volunteers filled in questionnaires about their health and lifestyle, first in 1975 and again in 1981. These included questions about height, weight and motivation to lose weight. Even after controlling for smoking and excluding anyone with a chronic illness that could have led to weight loss, Sørensen found that overweight or obese people who intended to lose weight in 1975 and succeeded were nearly twice as likely to have died by 1999 compared with those who had no intention to lose weight and stayed about the same (Public Library of Science Medicine, vol 2, e171).

Sørensen is not sure why weight loss should carry this extra risk, but he speculates that it has something to do with loss of lean tissue. He points to several studies showing that the lower your lean mass as a proportion of total mass the greater your risk of dying. "For some reason, to have a low or declining lean body mass is pretty dangerous, but why this should be so is not clear," he says.

His study was relatively small - only 2957 subjects - and it is too early to jump to conclusions. "I wouldn't want to say on the basis of our paper that anyone should stop trying to lose weight," he says.

Fortunately, there is lots of research showing that losing weight through exercise helps to preserve or increase lean mass. And Ludwig points to early evidence from animal studies that diets that cut out refined carbohydrates such as white bread may reduce fat mass while maintaining lean mass. "There's no reason why all diets should have the same impact on lean and fat mass," he says.

All this will be music to the ears of Stanton Glantz, who gave up sugary soft drinks years ago. His BMI may say he is borderline obese, but he reckons he's got plenty of lean mass and a robust heart. "That's a lot more important than what you look like," he says. If Flegal's work is to be believed, he can look forward to a long and healthy life.