Are restaurant calorie counts accurate?

On average yes, individually often not. When 269 restaurant items were measured by bomb calorimetry the overall difference from the stated figures was negligible, but 19% of items contained at least 100 kcal per portion more than stated, and the understatement was worst on items advertised as lower in energy.

How wrong is my own food log likely to be?

Pooled validation studies that compared self-reported intake with recovery biomarkers found average under-reporting of 28% with a food-frequency questionnaire and 15% with a single 24-hour recall. Logging closer to the moment of eating is what shrinks that gap.

If the numbers are wrong, why count at all?

Because the useful signal is the change, not the level. A repeated error behaves like a fixed offset and cancels when you compare one week with the next, which is why consistency of method matters more than accuracy of any single entry.

▸ Article

How accurate is calorie counting?

Last updated: 2026-09-22

Less accurate than the decimal places suggest, and accurate enough to be worth doing. Measured in a laboratory, 19% of restaurant items carried at least 100 kcal per portion more than the menu stated^[1], and people under-report what they ate by 15% to 28% on average^[4]. Counting still works because a deficit is judged on how your own number moves from week to week, not on whether it is true in absolute terms.

On this page

Where the error comes from

A logged meal passes through three estimates before it becomes a number. What the food actually contained. How much of it you actually ate. And what you actually remembered to record. Each of the three has been measured separately, and none of them is small.

They also do not cancel each other out. Two of the three have a direction: food tends to contain more than the label claims, and people tend to remember eating less than they did. Errors that share a sign accumulate rather than average away, which is how a day that looks like a deficit on paper can be maintenance in the body.

The number printed on the menu

The most careful test of restaurant figures burned the food. Researchers measured 269 items from 42 restaurants by bomb calorimetry and compared the result against the stated energy. Overall the menus were honest: the average difference was 10 kcal per portion, which is nothing. But the average hid the spread. Fifty of the 269 items, or 19%, contained at least 100 kcal per portion more than stated, and the worst tenth averaged 289 kcal per portion above their stated value^[1].

The bias was not random either. Items with low stated energy tended to contain more than stated, and items with high stated energy tended to contain less. The same group's earlier work found that reduced-energy restaurant dishes measured 18% above their stated values on average, and that free side dishes raised the energy actually served to 245% of what had been stated for the dish they arrived with^[2].

As a rule of thumb: the lighter an item claims to be, the more likely the claim is optimistic. The dishes you choose precisely because they are the careful option are the ones whose numbers deserve the least trust.

The portion is the larger variable

Most of what people eat in Egypt and the Gulf never had a number attached to it in the first place. There is no label on a pot of mahshi, on a tray of koshari from the shop downstairs, or on rice cooked at home. The name of the dish is fixed. Almost nothing else about it is.

Two things move a home-cooked plate more than anything else: the fat added at the end, and the size of the serving spoon. Rice finished with ghee and rice finished with water carry the same name in every food database and are not the same food. A cut of lamb the butcher trimmed and one he did not are not the same food. A shared plate is not a portion at all until somebody decides what the portion was.

This is the error most worth attacking, because it is the one you can genuinely reduce. A kitchen scale used on a handful of dense items removes more uncertainty in a week than any amount of hunting for a better database entry.

Memory is the largest error of all

The classic result came from people who were certain they were eating very little and losing nothing. Their energy expenditure turned out to be normal, within 5% of what their body composition predicted, so a slow metabolism was ruled out as the explanation. They had under-reported what they ate by 47% and over-reported their physical activity by 51%^[3].

That group was selected for diet resistance, so it is the extreme rather than the norm. The ordinary case is better, and still substantial. Pooling five validation studies that checked self-reported intake against recovery biomarkers, under-reporting of energy averaged 28% with a food-frequency questionnaire and 15% with a single 24-hour recall^[4].

Notice what separates those two figures. The recall asked about one recent day; the questionnaire asked people to summarise months. The further a record sits from the moment of eating, the more of the meal it loses. That is an argument about when you log, not about how honest you are.

The target you count against is an estimate too

The other side of the comparison carries error of its own. Resting metabolic rate is predicted from a formula, and in the systematic review that compared the formulas in common use, the best of them landed within 10% of the measured value more often than any other and had the narrowest error range^[5]. "Best available" and "correct for you" are different claims.

The activity multiplier applied on top of it is not a measurement at all. It is a category you picked off a list. The guide on how your daily calories are calculated sets out each step and what it assumes about you.

Why counting still works

Because you are not trying to learn the truth about a meal. You are trying to produce a number that moves in the right direction when your eating changes, so that this week can be compared with last week. An error that repeats is not really an error for that purpose. It is an offset, and it cancels when you subtract one week from the next.

That is also the test of which habits are worth the effort. Weighing the oil is worth it, because oil is dense, invisible once it is in the pan, and your guess about it is not stable. Arguing with a database entry for a vegetable is not. Logging the same dish the same way every time matters more than logging it correctly.

So: record at the moment of eating rather than at the end of the day. Weigh the few dense things — oil, rice, bread, nuts, cheese — and estimate the rest. Count the drinks, and the food eaten standing up in the kitchen, which are the two categories that vanish from every log. And keep your method identical from week to week, even when you suspect it is wrong.

Then let the trend correct the target. If three or four weeks at a given number produce no movement, the estimate was wrong for you, and the answer is to change the number rather than to count harder. The guide on how fast you can safely lose weight covers what the trend ought to look like and where the limits sit.

Sources

Try it on WhatsApp