Nutrition Labels Can Be Legally 20% Off: Food Log Math

A calorie label is a rounded estimate, and the rules leave real room around it. Under the US regulation 21 CFR 101.9(g)(5), a packaged food is only misbranded if its measured calories run more than 20 percent above the declared value — so a "200 kcal" pack can legally contain up to 240 kcal. Packaged snacks usually land close to their labels, but restaurant and freshly made meals can drift much further, which matters when you log them.

If you weigh your food to the gram and still feel like the scale is not moving the way the maths says it should, the label itself may be part of the gap. Nutrition panels look like measurements — a single tidy number per serve — but they are calculated estimates that regulators allow to vary within a defined band. Knowing how wide that band is, and which foods sit at the wide end, helps you read your own calorie tracking with the right amount of trust: enough to steer by, not so much that you chase a false decimal point.

This guide walks through what the tolerance rules actually permit, what the strongest measurement studies found when researchers put labelled foods in a bomb calorimeter, and how to build a small, honest error buffer into your log. Every figure below is grounded in a primary source — peer-reviewed calorimetry studies and the regulations themselves — listed at the end.

How accurate are nutrition labels legally required to be?

For calories and nutrients people limit — fat, saturated fat, sodium, sugars — US rule 21 CFR 101.9(g)(5) deems a food misbranded only if the measured content is more than 20 percent above the declared value. A label reading 200 kcal can legally hold up to 240 kcal. Australia and New Zealand apply a comparable 20 percent working tolerance under Standard 1.2.8.

The asymmetry is deliberate. For nutrients you might want more of — fibre, protein, vitamins, minerals — the measured amount has to be at least 80 percent of the label. For nutrients you might want less of, the measured amount cannot exceed 120 percent. The rule protects consumers in the direction that matters for each nutrient, but it also means the "calories" line is a ceiling-bounded estimate, not a precise readout.

Enforcement is light. The regulator does not routinely test products; compliance rests mostly with manufacturers, and testing is targeted rather than universal. That does not mean labels are routinely wrong — most packaged foods are close, as the data below shows — but it does mean the number is set by the maker, within a band, and rarely re-checked.

Do measured calories really differ from the label?

Yes, but the size of the gap depends heavily on the food. Packaged, mass-produced items track their labels closely: a 2013 bomb-calorimetry study of 24 US snack foods found a median excess of just 6.8 kcal, or 4.3 percent. Restaurant and freshly prepared meals drift far more — a 2010 study of 29 reduced-energy restaurant items measured them at 18 percent above stated values on average.

The difference comes down to control. A factory making millions of identical bars can calibrate its recipe tightly; a kitchen plating a burrito by hand cannot. Here is what the strongest measurement studies found when researchers combusted real foods and compared the heat released against the printed number.

The pattern is consistent across two decades and several countries: mass-produced packaged foods are usually within single-digit percentages, while anything assembled fresh — restaurant plates, café items, meal-kit dishes — carries a wider and less predictable error. That is the same reason calorie apps drift on vegetables and mixed dishes: the more a human hand shapes the portion, the more the true energy varies.

Which foods carry the biggest labelling error?

Freshly assembled and lower-calorie foods carry the biggest error. In the 2011 restaurant study, dishes marketed as low-calorie were the most understated, and the foods with the greatest excess averaged about 289 kcal above their stated value. Packaged, shelf-stable items with tight manufacturing sit at the accurate end, typically within 5 to 10 percent.

A useful mental model is to sort foods by how much variability sits between the recipe and your plate. The table below groups common foods by typical labelling reliability, drawn from the measurement studies above.

Australian research shows the same wide spread. A 2006 study of 350 samples across 70 products found individual nutrient declarations varying from 13 percent under to 61 percent over, and under a strict 20 percent tolerance only about 16 percent of products fully complied for every listed nutrient — though many misses were tiny amounts that round awkwardly rather than meaningful energy gaps.

How should you adjust your food log for label error?

Treat your daily total as a range, not a single number, and stop refining past about 10 percent. Log the label value as your best estimate, then hold a mental buffer — roughly 5 percent for packaged-food days and 10 to 15 percent for restaurant-heavy days — rather than trying to correct each item. Consistency over weeks beats precision on any single day.

Because the errors are partly random, they tend to average out across many items and many days. A week of logging is far more reliable than any one meal, so the practical move is to keep your method steady and read the trend. Here is a simple, non-fussy way to fold label tolerance into your tracking without turning every meal into a maths problem.

This is the same logic behind not double-counting exercise calories: when your inputs each carry meaningful uncertainty, stacking precise-looking numbers creates a false sense of exactness. Research on tracking apps points the same way — a 2026 review of app accuracy found database and portion errors of a similar scale to label tolerance, so the winning strategy is a steady method read over weeks, not a perfect day. If you want to see how the apps themselves stack up, the MyFitnessPal versus Cronometer accuracy review breaks that down.

None of this means tracking is pointless — far from it. A number that is consistently within 10 to 20 percent, logged the same way every day, is more than accurate enough to create a reliable deficit or surplus. The goal is a trustworthy trend, and the label gives you exactly that: a solid estimate to steer by, as long as you do not mistake it for a measurement.

Frequently Asked Questions

Can a food legally have more calories than its label says?

Yes. Under US rule 21 CFR 101.9(g)(5), a food is only considered misbranded if its measured calories exceed the declared value by more than 20 percent. A product labelled 200 kcal can legally contain up to 240 kcal. Australia and New Zealand apply a comparable 20 percent working tolerance under the Food Standards Code.

Are packaged food labels usually accurate?

Mostly, yes. A 2013 bomb-calorimetry study of 24 common US snack foods found a median excess of only 4.3 percent, and most products sat within the 20 percent legal allowance. Mass-produced packaged foods are calibrated tightly, so their labels are usually your most reliable number.

Why are restaurant calorie counts less accurate?

Because they are assembled by hand. A 2010 study of 29 reduced-energy restaurant items measured them at 18 percent above stated values on average, and a 2025 England audit found 34 percent of menu items fell outside the 20 percent accuracy threshold. Oil, sauce, and portion size vary plate to plate in ways a factory recipe does not.

Should I add a percentage to every calorie I log?

Not to every item. A simpler approach is to log packaged foods at their label value, add a modest buffer (roughly 15 to 20 percent) only on restaurant-heavy days, and watch your weekly average rather than any single day. Random label errors tend to cancel out over time.

Does label inaccuracy make calorie tracking pointless?

No. Even with a 20 percent tolerance, a number logged the same way every day gives a reliable trend, which is what drives a deficit or surplus. The practical limit is around 10 percent precision — chasing single-calorie accuracy on an input that is legally allowed to vary by a fifth is false precision, not better tracking.

Sources