The figure everyone repeats — that 95% of diets fail — has no paper behind it. It traces to a single 1959 study: a review of the then-existing literature, plus a series of 100 consecutive patients at the Nutrition Clinic of New York Hospital who were, in the practice of the day, handed a diet sheet and followed up (Stunkard 1959). The numbers in that work, in the author's own later account, are 12% losing 20 pounds, 1% losing 40 pounds, and 2% still holding a 20-pound loss two years later (Stunkard 1983). Nobody wrote 95%. Set against that: when someone defines success and measures it, roughly 20% of people with overweight achieve and hold a loss of at least 10% of body weight for at least a year (Wing 2005).
This article does two things. It traces the provenance of a number that has been circulating for six decades without a source, and it sets out what is genuinely well evidenced about why holding a weight loss is harder than achieving one.
Long-term weight-loss maintenance: the definition used in the research that reports a roughly 20% success rate — intentionally losing at least 10% of initial body weight and keeping it off for at least a year. Any success or failure percentage without a definition attached is uninterpretable, because the answer moves entirely with where the bar is set.
Where the number came from
Albert Stunkard and Mavis McLaren-Hume published "The results of treatment for obesity" in the AMA Archives of Internal Medicine in 1959. It has two halves: a review of eight earlier reports, and an original case series (Stunkard 1959).
The PubMed record for that paper carries no abstract, and its full text was not opened for this article. Every figure below therefore comes from a document that was read in full — Stunkard's own Citation Classic commentary on his paper, written in 1983 and archived at the University of Pennsylvania.
In his own words, on the literature half: "With the exception of Feinstein's own series, only 25 percent of patients lost more than 20 pounds and only five percent lost more than 40 pounds."
On his own series: "I enlisted the help of a dietician, Mavis McLaren-Hume, in following the course of 100 consecutive patients referred to the Nutrition Clinic of New York Hospital… These results were, as I had suspected, even worse than those in the literature. Instead of 25 percent, only 12 percent of patients lost 20 pounds, and instead of five percent, only one percent lost 40 pounds. Furthermore, two years later, only two percent of patients had maintained their 20-pound weight loss" (Stunkard 1983).
Twenty pounds is about 9 kg; forty pounds about 18 kg.
Now look for 95% in any of that. It is not there. What is there is 5% and 2% and 1% and 12% and 25% — and 95 is what you get by subtracting one of them from a hundred, rounding, and dropping the denominator, the threshold, the follow-up period and the era it belonged to. A page that publishes "95% of diets fail" is quoting an arithmetic complement of a 1950s clinic series with everything that made it interpretable removed.
What that study could and could not measure
Stunkard's paper was, on his own assessment, important: "I believe that this paper has been cited frequently because it documented for the first time the ineffectiveness of outpatient treatment for obesity." By 1983 it had been cited in more than 305 publications.
But look at what "outpatient treatment for obesity" meant in the 1950s. One hundred self-selected referrals to a hospital nutrition clinic. No control group. No randomisation. A diet sheet as the intervention. This was before behavioural treatment, before structured maintenance programmes, before pharmacotherapy of any modern kind, and before bariatric surgery. Its finding — that handing people a diet sheet did not work well — is a real finding about handing people a diet sheet.
Carrying that number forward to describe what happens to a person in 2026 requires assuming that nothing has changed in sixty-seven years, which the intervening literature does not support.
What the success rate looks like when someone defines it
The corrective to a decontextualised failure rate is a defined success rate.
A 2005 review opens with exactly the problem this article is about: "There is a general perception that almost no one succeeds in long-term maintenance of weight loss. However, research has shown that approximately 20% of overweight individuals are successful at long-term weight loss when defined as losing at least 10% of initial body weight and maintaining the loss for at least 1 y" (Wing 2005). An earlier synthesis by the same research tradition puts the figure above 20% on the same definition, and adds that once a loss has been held for two to five years, the odds of longer-term success rise substantially (Wing 2001).
Two things about that 20%. It rests entirely on its definition — move the threshold to 20% of body weight, or the duration to five years, and the number changes. And it is a population figure, not a probability that applies to any individual person — individual results vary. Both reviews predate the current medication era.
The most rigorous long-term look at intensive lifestyle treatment lands in between. The Look AHEAD trial, one of the largest and longest lifestyle-intervention studies ever run, produced a mean weight loss of about 4.7% at eight years — a figure that reflects substantial rebound from higher early losses (Look AHEAD 2014). The pattern is neither total failure nor durable success. It is a real average loss that is smaller than what was achieved in year one.
The physiology: energy expenditure falls further than it should
This is the first of two measured reasons that holding a loss is harder than producing one.
In a controlled weight-perturbation study, maintaining body weight 10% or more below usual weight reduced total energy expenditure by more than the change in body composition alone predicts. The effect appeared in people who had never been obese as well as in people with obesity (Leibel 1995). In other words, a reduced-weight body spends less energy than an equivalent body that was always that size.
How long that persists is best illustrated by the most extreme dataset available, with its limits stated. Fourteen participants from a televised weight-loss competition lost a mean 58.3 kg. Six years later they had regained a mean 41.0 kg, and their resting metabolic rate was still 704 kcal/day below baseline — with metabolic adaptation, after adjusting for body composition and age, running at −499 kcal/day (Fothergill 2016). The authors describe the adaptation as "a proportional, but incomplete, response".
Fourteen people, a 58 kg mean loss, and televised competition conditions. Those figures do not transfer quantitatively to ordinary or medically supervised weight loss, and the authors' own word — "incomplete" — cuts against the popular version, which is that dieting permanently breaks metabolism. A review by the same researchers sets out the proposed mechanism: reduced body weight is defended by a coordinated set of metabolic, neuroendocrine and behavioural responses (Rosenbaum 2010).
The physiology: hunger does not reset
The second measured reason is appetite, and the evidence is unusually direct.
Fifty adults with overweight or obesity and without diabetes completed a ten-week very-low-energy diet, losing a mean 13.5 kg. Hormones were measured at baseline, at ten weeks, and again at 62 weeks. At the end of the diet, leptin, peptide YY, cholecystokinin, insulin and amylin were all significantly reduced, ghrelin and two other appetite-related hormones were significantly increased, and subjective appetite had risen significantly.
A year after the initial loss, every one of those differences was still significant — including hunger (Sumithran 2011). The authors' conclusion, verbatim: "levels of the circulating mediators of appetite that encourage weight regain after diet-induced weight loss do not revert to the levels recorded before weight loss. Long-term strategies to counteract this change may be needed to prevent obesity relapse."
The study has no control group, n=50, and used one very-low-energy protocol, so no percentage change should be attached to any individual hormone. What it establishes is the direction and the persistence: a year on, the body is still signalling for the weight to come back.
That is the answer to the moral framing of dieting. Someone rebounding after a loss is contending with a measurably altered energy expenditure and a measurably altered appetite drive, both of which are still present twelve months later.
What the people who hold a loss actually do
The National Weight Control Registry recruits people who have already succeeded, which makes it useful for description and useless for proof.
Its founding cohort of 784 members had each lost at least 13.6 kg and kept it off for five years, with a mean loss of 30 kg. They reported a low-fat, low-calorie diet and a high level of physical activity — roughly an hour a day (Klem 1997). A later analysis of 3,003 registry members found that how often people weighed themselves tracked what happened over the following year: those who reduced their weighing frequency regained 4.0 ± 6.3 kg, those who kept it unchanged regained 1.8 ± 5.3 kg, and those who increased it regained 1.1 ± 6.5 kg (Butryn 2007).
Every one of those findings is observational, self-reported, and drawn from a self-selected group of people who already did well. Survivorship bias is built into the design. They describe what maintainers do; they do not establish that doing those things causes maintenance. We cover this in more depth in the habits that predict keeping it off.
How to read a "diets don't work" claim
Three questions strip most of these claims down:
- What counted as failure? Losing less than 20 pounds is a different bar from regaining more than half of what was lost.
- Over what period, and measured how? Two years in a 1950s clinic, eight years in a randomised trial and one year in a registry are not the same measurement.
- Compared with what? A failure rate quoted with no comparator is a rhetorical device rather than a statistic.
The physiology above is what makes maintenance hard, and it is well evidenced. The 95% figure is not, and it never was. If you want to understand what is happening in your own case rather than in a rounding from 1959, that starts with a clinical assessment — you can check your eligibility and speak to a doctor about what would be suitable.
Common questions
Where does the 95% diet failure statistic come from?
From a rounding, not a study. It traces back to a 1959 paper by Stunkard and McLaren-Hume — a review of the existing literature plus a series of 100 consecutive patients at one New York hospital clinic (Stunkard 1959). In Stunkard's own later account, that series found 12% losing 20 pounds, 1% losing 40 pounds, and 2% still holding a 20-pound loss two years later (Stunkard 1983). No source in that chain states 95%.
So do most diets fail?
It depends entirely on the definition, which is why an unqualified number is not useful. Using a defined standard — losing at least 10% of initial body weight and holding it for at least a year — roughly 20% of people with overweight succeed (Wing 2005). Over eight years, an intensive lifestyle programme produced a mean 4.7% weight loss, well below its own early results (Look AHEAD 2014).
Why is keeping weight off so much harder than losing it?
Two measured changes work against it. Energy expenditure falls by more than the change in body mass alone predicts, in people who were never obese as well as in people who were (Leibel 1995). And a year after a 13.5 kg loss, ghrelin was still elevated, leptin and peptide YY still reduced, and subjective hunger still significantly higher than before the loss (Sumithran 2011).
Does metabolism stay slower permanently?
The most extreme dataset says it can persist for years, and it is a small and unusual one. Fourteen participants from a televised weight-loss competition still had resting metabolic rates several hundred kcal a day below what their body composition predicted six years later (Fothergill 2016). Mean weight loss in that group was 58 kg under competition conditions, so the magnitude does not transfer to ordinary or medically supervised weight loss.
What do people who keep weight off actually do?
Registry data describe a consistent pattern: around an hour a day of physical activity and a low-calorie, low-fat diet (Klem 1997), with regular self-weighing — people who weighed themselves less often over a year regained about 4.0 kg against 1.1 kg among those who weighed more often (Butryn 2007). These registries are self-selected and self-reported, so they describe what maintainers do rather than proving what causes maintenance.