# How accurate are restaurant calorie counts?

July 16, 2026 · Calorie Tracking · 8 min read · https://burnweek.fit/blog/how-accurate-are-restaurant-calorie-counts/

> The menu is most wrong exactly where you trust it most — and the salad you ordered for its number is the likeliest item on the page to beat it.

**Key takeaways**

- Menus are honest on average and unreliable per plate: across 269 items the mean gap was just 10 kcal, yet 19% of items ran 100+ kcal over stated (Urban 2011).
- The error tilts by dish — low-calorie items measured higher than stated while high-calorie items measured lower, so the salad you ordered for its number is the likeliest to beat it.
- The same meal at different branches differed more than twofold — 1,386 kcal versus 657 kcal for one Chinese dish across 111 restaurants (Roberts 2018).
- Menu calories needn't come from a lab: US rules allow "nutrient databases..., cookbooks, laboratory analyses, or other reasonable means", and only chains with 20+ locations must post at all.
- Sit-down beats fast food for calories, not against it: full-service meals averaged 1,317 kcal versus 809 for fast food, with 94% over 600 kcal.

## Right on average, unreliable per plate

How accurate restaurant calorie counts are depends entirely on whether you mean the menu or your dinner. When researchers bomb-calorimetered 269 food items from 242 unique foods across 42 restaurants, the stated calories were, in aggregate, essentially correct — the measured-versus-stated difference came to just 10 kcal per portion (95% CI, −15 to 34) ([Urban et al., 2011](https://pubmed.ncbi.nlm.nih.gov/21771989/)). Menus are not systematically lying.

And that average is close to useless to you, because you do not eat the average. In the same study, 50 items — 19 percent — contained at least 100 kcal *more* per portion than the menu claimed, and when the worst offenders were resampled they still measured 289 and 258 kcal above stated ([Urban et al., 2011](https://pubmed.ncbi.nlm.nih.gov/21771989/)). The errors cancel across a dataset. They do not cancel across your dinner. A menu number tells you roughly what a dish of that kind contains; it says much less about the one the kitchen just handed you.

## The direction of the error is not random

Buried in that JAMA analysis is the finding that should change how you read a menu. The errors were not scattered — they tilted by dish. Foods with **lower** stated energy contents measured **higher** than stated, while foods with higher stated energy measured lower ([Urban et al., 2011](https://pubmed.ncbi.nlm.nih.gov/21771989/)). The salad understates. The burger overstates.

> The menu is most wrong exactly where you are trusting it most: the low-calorie item you ordered *because* of its number is the one likeliest to exceed it.

The "light menu" evidence sharpens this. Testing reduced-energy items specifically, measured values averaged 18 percent above stated for 29 restaurant foods, individual items reached 200 percent of their stated value, and free side dishes pushed total provided energy to an average of 245 percent of the entrée's stated calories ([Urban et al., 2010](https://pubmed.ncbi.nlm.nih.gov/20102837/)). That last number is the one people miss: the bread, the fries that came with it, the dressing. The entrée was fine. The meal was not — which is a recurring theme in [hidden calories stalling weight loss](/blog/hidden-calories-stalling-weight-loss).

The authors pressed one caveat, and it strengthens the case rather than weakening it: those differences substantially exceeded laboratory measurement error but did not reach statistical significance, because the degree of underreporting varied so much from item to item ([Urban et al., 2010](https://pubmed.ncbi.nlm.nih.gov/20102837/)). The instability *is* the finding. A menu's problem is not a fixed bias you could learn to subtract — it is that the size of the error is itself unpredictable, which is precisely what a single confident number cannot express.

## The same dish is a different dish at the next branch

Menu accuracy assumes the dish is reproducible. It often isn't. A multi-country study measured 233 meals from 111 randomly selected restaurants across six countries and found that identical meals bought at different locations differed by **more than twofold** — one Chinese dish measured 1,386 kcal at one outlet and 657 kcal at another ([Roberts et al., 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC6290458/)).

No menu number survives that. It isn't a labeling failure; it's a manufacturing one. A factory portions to the gram, a cook portions by hand and by mood.

| Restaurant type | Mean measured energy | Share ≥600 kcal |
| --- | --- | --- |
| Full service | 1,317 kcal (SD 442) | 94% |
| Fast food | 809 kcal (SD 363) | 72% |

Data: [Roberts et al., 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC6290458/). Note which row is larger. The sit-down restaurant delivered 33 percent *more* energy per meal than fast food — the opposite of most people's intuition, and a reminder that [portion distortion](/blog/portion-distortion-restaurant-sizes) is not a fast-food-only phenomenon.

## Where a menu number actually comes from

This is the part almost nobody knows, and it explains everything above. US menu labeling applies only to a restaurant "that is a part of a chain with 20 or more locations doing business under the same name...and offering for sale substantially the same menu items" ([21 CFR 101.11](https://www.law.cornell.edu/cfr/text/21/101.11)). Most independent restaurants post nothing, and are not required to.

For those that are covered, the calorie figure need not come from a laboratory at all. The regulation states that "nutrient values may be determined by using nutrient databases (with or without computer software programs), cookbooks, laboratory analyses, or other reasonable means" ([21 CFR 101.11](https://www.law.cornell.edu/cfr/text/21/101.11)). A menu calorie count can legally be a recipe calculation — a spreadsheet of intended ingredients in intended amounts.

Which makes the compliance mechanism clear. The rule requires that the method of preparation and the amount of a standard menu item offered for sale "adhere to the factors on which its nutrient values were determined" ([21 CFR 101.11](https://www.law.cornell.edu/cfr/text/21/101.11)). In other words, the number is accurate *if the kitchen makes the dish exactly as specified*. It is a statement about the recipe, not a measurement of the plate. A generous hand with the oil doesn't make the menu wrong in any legal sense — it makes your dinner different from the dish the number describes. Contrast this with packaged food, where [nutrition labels](/blog/how-accurate-are-nutrition-labels) carry an enforceable 20 percent tolerance against laboratory measurement of the actual product.

## The unlabeled restaurant: worse, and not for the reason you think

So what about the independent place with no numbers at all? Researchers randomly sampled 364 meals from 123 non-chain restaurants across three US cities and measured them by bomb calorimetry. The meals averaged 1,205 ± 465 kcal, ranged from 113 to 3,008 kcal, and 92 percent exceeded typical energy requirements for a single eating occasion ([Urban et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5746190/)).

Two results from that study deserve more attention than they get. First, non-chain meals were **not significantly different** from equivalent large-chain meals (+5.1%, P=0.41) — the independent kitchen is not quietly healthier, it is merely unmeasured. Second, and more useful: USDA database values sat only about 2.3 percent below the measured energy, a non-significant gap ([Urban et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5746190/)). Estimating an unlabeled restaurant meal from a decent database is a genuinely reasonable move. It performed about as well as the menu would have.

## What to actually do with a menu number

Use it, and widen it. Concretely: log the menu figure but treat it as the *floor* of a band rather than the center, especially for anything marketed as light — that is where the understatement concentrates. Count the sides separately, since they were never in the entrée's number. Assume a sit-down meal is bigger than it looks; the measured average for full-service meals was 1,317 kcal ([Roberts et al., 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC6290458/)). And when there's no menu figure at all, estimate from a database without guilt — it lands about as close as the posted number would have ([Urban et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5746190/)).

This is precisely the situation a single confident number handles worst. A restaurant meal is genuinely ambiguous — the kitchen improvised, the portion was hand-scooped, the oil is invisible — and an estimate that admits it by reporting "700 to 1,100" is telling you something true that "890" conceals. Read a wide band for what it is — a signal to be conservative today rather than to trust the digit. More on the general principle in [why calorie counts are ranges](/blog/why-calorie-counts-are-ranges), and on the full stack of error in [how accurate calorie counting is](/blog/how-accurate-is-calorie-counting). For the tactics, see [estimating restaurant meal calories](/blog/estimating-restaurant-meal-calories).

## FAQ

### Are restaurant menu calories legally required to be accurate?

Only loosely. Menu labeling applies to chains with 20 or more locations, and nutrient values may be derived from "nutrient databases..., cookbooks, laboratory analyses, or other reasonable means" ([21 CFR 101.11](https://www.law.cornell.edu/cfr/text/21/101.11)). The declaration describes the standardized recipe, and compliance rests on the kitchen preparing the dish to the factors the values assumed — not on testing the plate you were served.

### Should I add a buffer to restaurant calorie counts?

For low-calorie and "light" items, yes. Measured energy for reduced-energy restaurant foods averaged 18 percent above stated, with some items at 200 percent of their claim ([Urban et al., 2010](https://pubmed.ncbi.nlm.nih.gov/20102837/)), and the JAMA data showed low-stated dishes running high while high-stated dishes ran low ([Urban et al., 2011](https://pubmed.ncbi.nlm.nih.gov/21771989/)). Buffer the salad, not the steak.

### Is a sit-down restaurant a safer choice than fast food?

Not on calories. Measured full-service meals averaged 1,317 kcal against 809 kcal for fast food — about 33 percent more — and 94 percent of full-service meals exceeded 600 kcal ([Roberts et al., 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC6290458/)). Non-chain restaurants also matched large chains for energy content ([Urban et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5746190/)).

## Sources

- [Urban LE, et al. Accuracy of stated energy contents of restaurant foods. JAMA. 2011.](https://pubmed.ncbi.nlm.nih.gov/21771989/)
- [Roberts SB, et al. Measured energy content of frequently purchased restaurant meals: multi-country cross sectional study. BMJ. 2018.](https://pmc.ncbi.nlm.nih.gov/articles/PMC6290458/)
- [Urban LE, et al. Energy contents of frequently ordered restaurant meals and comparison with human energy requirements and USDA database information: a multisite randomized study. J Acad Nutr Diet. 2016.](https://pmc.ncbi.nlm.nih.gov/articles/PMC5746190/)
- [Urban LE, et al. The accuracy of stated energy contents of reduced-energy, commercially prepared foods. J Am Diet Assoc. 2010.](https://pubmed.ncbi.nlm.nih.gov/20102837/)
- [21 CFR 101.11 — Nutrition labeling of standard menu items in covered establishments. Code of Federal Regulations (Cornell Legal Information Institute).](https://www.law.cornell.edu/cfr/text/21/101.11)