Guide

How close is close enough?

Calorie targets look exact. The numbers feeding them are not. Labels are rounded, they are legally allowed to understate, restaurant counts are estimates, and some cuts lose nearly a third of their weight in the pan. Here is how much error a normal day carries, and which parts are worth fixing.

By Alen Yaco · Double bachelor's in biological sciences and social ecology, UC Irvine · Published 18 September 2026

Not on this page. Working out the target itself is not on this page, that arithmetic lives on /calorie-calculator/, and how many grams of protein to aim for lives on /protein-calculator/.

How accurate does calorie counting need to be?

Calorie counting needs to be consistent more than exact. Most of the error in a logged day repeats itself day after day, so it cancels out of the trend. Your target is an estimate, and the numbers you log against it are estimates too. Judge both over weeks.

Label rounding, legal tolerance, restaurant estimation, raw against cooked weight, and the database row you picked: five things sit between the food and the number in your diary, and every one is larger than the decimal place people worry about. A sixth, the food that never reached the diary, has its own page.

What the five share is that they repeat. Yogurt logged 15 calories low is 15 calories low every morning, so the week-to-week change in your total is untouched. Errors that shift cost far more than errors that hold steady.

The target carries error too: an equation fitted to a population, then applied to one person. Producing it is the job of the calorie calculator.

How far off can a nutrition label legally be?

A US nutrition label may not understate calories by more than 20 percent. There is no matching rule in the other direction. A food is misbranded if laboratory analysis finds the composite more than 20 percent above the declared calories. Shortfalls are accepted under good manufacturing practice.

Labels are widely said to be allowed 20 percent either way. They are not. The ceiling in 21 CFR 101.9(g)(5), from the US Food and Drug Administration, runs one way: a food is misbranded if the composite holds more than 20 percent in excess of the declared value. Paragraph (g)(6) handles the other direction in one sentence with no percentage in it. The test runs on a composite from a production lot, not on your package.

The 80 percent people half-remember belongs to another list. Paragraph (g)(4)(ii) requires Class II nutrients, vitamins, minerals, protein, carbohydrate and fibre, to reach 80 percent of the declared value. Protein therefore has a legal floor. Calories do not.

Why the printed number is always round

Before any of that, the figure is rounded by law. Paragraph (c)(1) requires the nearest 5 calories up to and including 50, the nearest 10 above 50, and allows anything under 5 calories to be declared as zero.

So two foods 4 calories apart can print the same number, and two 3 apart can print numbers 10 apart. A spray oil used 10 times a day can carry close to 50 calories declared as zero. Protein rounds to the nearest gram, and the protein calculator sets that target.

Can I trust the calories printed on a restaurant menu?

Restaurant menu calories are estimates held to a looser standard than packaged labels. A covered chain needs a reasonable basis for the number, with no numeric tolerance attached. Measured in a laboratory, stated numbers were accurate on average, but individual dishes varied a great deal.

Under 21 CFR 101.11(c)(1), a chain may set its numbers from nutrient databases, cookbooks, laboratory analyses or other reasonable means. No 20 percent test, no composite sampling. Urban and colleagues, in JAMA in 2011, measured 269 items from 42 restaurants in a bomb calorimeter.

• Across all 269 items, measured energy differed from stated by 10 calories per portion, 95 percent confidence interval minus 15 to 34, P = .52. On average the menus were right.
• 50 items, 19 percent, held at least 100 calories per portion more than stated.
• Among sit-down restaurant entrees, not across all 269 items, foods with lower stated energy tended to hold more than stated, at P less than .001.

That last line is the useful one. In a sit-down restaurant, choosing the lightest thing on the menu means choosing from the group most likely to hold more than it claims. This is one 2011 study, most items measured once.

Should I weigh food raw or cooked?

Weigh food in whichever state the database entry describes, and stay with that choice. Raw and cooked weights are not interchangeable, because meat sheds water and fat in the pan while grains take water on. USDA measured broiled medium fat beef patties at a 69 percent yield, so 100 grams raw leaves 69 cooked.

USDA defines cooking yield as the cooked hot weight divided by the raw weight, and publishes it per cut and method in its Table of Cooking Yields for Meat and Poultry, Release 2. Across 175 items yields run from 29 to 96 percent, median 76. Every figure in the table below, including the sample counts and ranges, comes from that release and its data file.

Put 150 grams of cooked patty on the scale, log it against a raw entry, and at 69 percent yield you have logged 150 grams where 217 went into the pan. Weigh raw against a cooked entry and the diary reads high.

The chicken row is the weakest: the data file gives it no sample size and no standard deviation, and dates it to a 1975 release, so its 68 to 84 percent range says more than the midpoint. Even the well-sampled center loin chops ran from 58 to 84. A yield table shows the size of the effect, not what your pan did.

Grains move the other way

Meat loses water in the pan. Pasta and rice take it on, which moves calorie density further than almost anything else on the plate. The figures below are from USDA FoodData Central, SR Legacy, FDC IDs 169736, 169737, 168877 and 168878. Unenriched entries carry identical figures, so enrichment is not the variable. Water is.

Weigh 300 grams of cooked pasta against the cooked entry and you get 474 calories. Log the same 300 grams against the dry entry and you get 1,113, an error of 639 calories in one meal, larger than any meat yield can produce.

Dividing those densities suggests a cooked weight of roughly 2.3 times dry for pasta and 2.8 for rice, though USDA publishes no cooked yield for either grain and some starch leaves in the water. Weighing grains dry removes the problem.

Why do two entries for the same food show different calories?

Two database entries disagree because they usually describe different foods. Raw against cooked, a brand's own label against a generic laboratory analysis, per 100 grams against per serving. Label figures were rounded before they reached any database, and generic entries come from sampled foods, which vary.

• Different state. One row is raw or dry, the other cooked, the gap the section above works through.
• Different basis. Per 100 grams against per serving, and the serving is the manufacturer's choice.
• Different origin. Packaged rows come from labels, already rounded and allowed to hold 20 percent more than declared. Generic rows come from laboratory analysis of sampled foods, and samples differ.
• Different recipe. Two brands of one food are two foods.

The fix is not to hunt for the true entry. For a food you eat often, pick the row that matches how you weigh it and keep it. A row sitting 8 percent low leaves the diary 8 percent low every day, and the week-to-week comparison still works. Switching rows breaks that: the number moves while your eating does not.

Does one day over the target matter, or is the weekly average what counts?

One day over a calorie target moves almost nothing on its own. The weekly average is the number that matters, because that is what a weight trend responds to. A week swinging from 1,700 to 2,600 calories can still average 2,057, within 3 percent of a 2,000 target.

No day landed on the target and Saturday was 600 over, yet the week finished 400 calories above plan, an average of 57 a day. Hall, in the International Journal of Obesity in 2008, puts the planning rule at 32.2 megajoules per kilogram of weight lost, roughly 7,700 calories, the metric form of 3,500 per pound. Three warnings come with it.

• It describes weight lost, not fat. The same paper puts body fat at 39.5 megajoules per kilogram, roughly 9,441 calories, since weight lost includes lean tissue and glycogen with its water.
• It fits people with initial body fat above 30 kilograms and overestimates what leaner people need.
• Run forward it assumes a static deficit, so it predicts loss that never plateaus.

Put that rule over the week above and its 400 calorie overshoot is about 50 grams, less than a bathroom scale's swing between two mornings. The day is noise. The week is the signal, as long as the days you would rather skip are in it.

For a projection that allows for the body adapting, use the Body Weight Planner from the National Institute of Diabetes and Digestive and Kidney Diseases.

What is worth weighing, and what is not?

Weigh the foods that dominate the day and estimate the rest. Calorie-dense staples, oils, nut butters, grains and meat carry most of the error, because a small weighing mistake there is a large calorie mistake. Vegetables can be judged by eye. Consistency of method beats precision in one meal.

1. Match the state to the entry. Getting it wrong is 67 grams of beef on one 150 gram portion.
2. Weigh grains dry. The pasta example above is a 639 calorie swing on one plate.
3. Weigh by calorie density. The table below is the whole rule.
4. Keep one entry per repeat food. A steady error cancels out of the trend. A changing one does not.

Tare the bowl on the scale so the container never counts. Work in grams, because the same cup of rice or flour holds different weights depending on how it was packed. For oil, stand the pan on the scale and pour, since a poured tablespoon is not a measured one. Coffee, tea and leafy vegetables do not repay a scale.

When the scale and the diary disagree for a month, the first suspect is not your weighing. It is the target, fitted to people in general, and worth adjusting.

No logging method removes these error sources. A photo cannot see oil absorbed into the food, a barcode inherits the label's rounding, and a scale cannot report what a restaurant kitchen did. That is why CalorieCounter, which is not released yet and cannot be installed, marks a photo estimate Pretty sure, Best guess or Check this instead of printing a decimal it cannot support.

FAQ

Is the 20 percent label tolerance really plus or minus 20 percent?

No. For calories the cap runs one way. FDA guidance states it as a ratio: the laboratory result divided by the label figure must be 120 percent or less, and its worked example fails a product at 133 percent. Nothing sets a floor for calories, so plus or minus 20 percent is a misreading.

Do I need a kitchen scale to count calories properly?

Not for everything. The biggest improvement is free: match the state of the food to the entry, raw with raw and dry with dry. After that a scale earns its place on oils, nut butters, grains and meat, where 20 grams is a real number of calories.

My diary says I am in a deficit and my weight is not moving. What is wrong?

Usually nothing is wrong with the effort. Check the mechanics first: cooked food logged against raw entries, grains weighed after cooking, restaurant meals, and unfinished days. Then question the target: it came from a population equation, and the 7,700 calories per kilogram rule overestimates what leaner people need.

Sources

• US Food and Drug Administration, 21 CFR 101.9, Nutrition labeling of food (Code of Federal Regulations)
• US Food and Drug Administration, 21 CFR 101.11, Nutrition labeling of standard menu items in covered establishments (eCFR)
• US Food and Drug Administration, Guidance for Industry: Guide for Developing and Using Data Bases for Nutrition Labeling (1998)
• Urban LE and colleagues, Accuracy of stated energy contents of restaurant foods, JAMA 2011;306(3):287-93
• USDA Agricultural Research Service, USDA Table of Cooking Yields for Meat and Poultry, Release 2 (2014)
• USDA Table of Cooking Yields for Meat and Poultry, Release 2, data file
• USDA FoodData Central, SR Legacy, FDC IDs 169736, 169737, 168877, 168878
• Hall KD, What is the required energy deficit per unit weight loss?, International Journal of Obesity 2008;32(3):573-576
• Hall KD and Chow CC, Why is the 3500 kcal per pound weight loss rule wrong?, International Journal of Obesity 2013;37(12):1614
• National Institute of Diabetes and Digestive and Kidney Diseases, Body Weight Planner