This is not medical advice. General information about how food energy is calculated, not a treatment plan or a reason to eat less. If you are under 18, pregnant or breastfeeding, have ever had an eating disorder or a difficult relationship with food, or are managing a medical condition, talk to a doctor or a registered dietitian before you change what you eat. The full list is here.
Most of what gets written about tracker accuracy, including what I have written, treats the calorie count of a food as the fixed thing and the log as the part that misses it. Search the right entry, weigh the portion properly, and you have the number. That is the wrong way round for some foods, and researchers at the USDA have spent over a decade measuring how wrong.
The calorie value of a food is not measured. It is calculated, using three digestibility averages settled in the 1890s and applied to everything since. When somebody finally checked those averages against what actually leaves the body, whole almonds came back roughly a quarter below the calculated figure.
Two neighbouring questions have their own posts and this one deliberately leaves them alone. Every other link in the chain between your plate and your screen, and the published error on each, is in how accurate are calorie tracking apps. Which weight to put on the scale and which entry to pair it with is raw or cooked. This one is about the number those posts both take as given.
Where the 4, the 4 and the 9 come from
Every calorie figure you have ever logged traces back to one laboratory in Connecticut. The FAO's own account of the system is blunt about its age: the Atwater general factor system "was developed by W.O. Atwater and his colleagues at the United States Department of Agriculture (USDA) Agricultural Experiment Station in Storrs, Connecticut at the end of the nineteenth century (Atwater and Woods, 1896)."10
The method was to burn food in a calorimeter to get its heat of combustion, then correct that downward "for losses in digestion, absorption and urinary excretion of urea". What survived the corrections were three round numbers: 17 kJ/g (4.0 kcal/g) for protein, 37 kJ/g (9.0 kcal/g) for fat and 17 kJ/g (4.0 kcal/g) for carbohydrate, with alcohol added later at a rounded 29 kJ/g (7.0 kcal/g).10
Notice what those corrections are. They are population averages of how much of a nutrient a human digests and absorbs, measured on mixed diets more than a century ago. The system does not ask what food the protein arrived in or what it was wrapped in on the way down. The FAO says as much about why it lasted: "The Atwater system has been widely used, in part because of its obvious simplicity."10
A refinement exists and has done since 1955, when Merrill and Watt published factors that vary by food group, on the argument that "there are ranges in the heats of combustion and in the coefficients of digestibility of different proteins, fats and carbohydrates, and these should be reflected in the energy values applied to them."10 Nuts and dry legumes get 3.47 kcal/g for protein, 8.37 for fat and 4.07 for carbohydrate rather than 4, 9 and 4.10 Hold on to that, because it means the nut figure in a good database already carries a correction. It is still nowhere near enough.
One more quirk of the original worth knowing if you track fibre. "As originally described by Atwater, carbohydrate is determined by difference, and thus includes fibre."10 Fibre is largely fermented rather than digested, and the FAO notes that a factor of 8.0 kJ/g (2.0 kcal/g) for fibre was recommended in 1998 but "has not yet been implemented", the 2.0 resting on an assumption that fibre is 70 percent fermentable.10 So in the plain Atwater system, a gram of bran and a gram of sugar are the same four calories.
Almonds: 129 calories where the packet says 170
In 2012 a USDA team stopped arguing about the factors and ran the balance study. Eighteen healthy adults ate a fully controlled diet for 18 days in a crossover design, at one of three almond doses: 0, 42 or 84 g a day. For the final nine days of each period they collected all urine and all faeces, and the diets, the urine and the faeces were burned in a bomb calorimeter. Subtract what came out from what went in and you have the metabolisable energy of the almonds, measured rather than assumed.1
The answer was 4.6 ± 0.8 kcal/g, "which is equivalent to 129 kcal/28-g serving". The Atwater calculation for the same almonds gives 6.0 to 6.1 kcal/g, or 168 to 170 kcal a serving. In the authors' summary, "the Atwater factors, when applied to almonds, resulted in a 32% overestimation of their measured energy content."1
That plus or minus 0.8 is worth as much attention as the 4.6. It is a wide spread on eighteen people, and it is the reason this is a finding about almonds rather than a number you can apply to your afternoon.
The same laboratory has since done walnuts and cashews with the same method. Walnuts: 18 adults, mean age 53.1 years, two three-week feeding periods, 42 g a day. One 28 g serving came out at 146 kcal against a calculated 185, which the authors put at "21% less than that predicted by the Atwater factors".3 Cashews: 18 adults, four-week periods, 42 g a day. A 28 g serving measured 137 kcal, "16% lower" than the labels, with energy digestibility of 92.9% on the cashew diet against 94.9% on the control.4
| Per 28 g serving | Calculated | Measured | As the authors put it |
|---|---|---|---|
| Almonds1 | 168 to 170 kcal | 129 kcal | 32% overestimation |
| Walnuts3 | 185 kcal | 146 kcal | 21% less than predicted |
| Cashews4 | about 163 kcal | 137 kcal | 16% lower than labels |
Those three percentages are not on the same footing and you should not line them up and average them. The almond paper divides the gap by the measured value, so 32% is how far the calculation sits above the truth. The walnut paper divides by the predicted value, so 21% is how far the truth sits below the calculation. Run the almond numbers the walnut way and you get 24%, not 32%. Same finding, two conventions, and anybody quoting one figure next to the other without saying which is quietly comparing different things.
The mechanism is physical rather than metabolic. Fat in a whole nut sits inside intact plant cell walls, and a good deal of it never meets a lipase before it leaves. Somebody measured that too, in 63 healthy adults in Ghana, Brazil and the USA with an average BMI of 21.8, fed whole peanuts, peanut butter, peanut oil or peanut flour. Faecal fat on whole peanuts was 5.22% against 3.07% on the oil, 3.11% on the butter and 3.75% on the flour, and faecal energy per gram followed the same order.6
For reference, USDA FoodData Central currently lists almonds at 579 kcal per 100 g, which works out at about 162 kcal for a 28 g serving.12 That is the Merrill and Watt nut-specific figure rather than plain 4/9/4, and it is still 26% above what the balance study measured.
Chop the same almond and the number goes up
If the fat is locked in cell walls, then breaking the walls should release it, and the size of that effect has been measured in the same design. A five-period crossover, 18 healthy adults, 42 g a day of almonds in one of four forms plus a control period with none, nine days adapting and nine days collecting everything.2
| Form of almond | Measured energy | Per 28 g |
|---|---|---|
| Whole, natural | 4.42 kcal/g | 124 kcal |
| Whole, roasted | 4.86 kcal/g | 136 kcal |
| Chopped | 5.04 kcal/g | 141 kcal |
| Almond butter | 6.53 kcal/g | 183 kcal |
Whole, roasted and chopped almonds all measured significantly below the Atwater prediction. Almond butter did not: at 6.53 kcal/g it was statistically indistinguishable from the calculated value, P = 0.08.2 Grind the nut finely enough and the convention becomes true again, because there is no longer a cell wall standing between the fat and your gut.
The researchers went looking for the physical reason and found it in a texture analyser. Roasted almonds took 298 ± 1.3 N to fracture against 345 ± 1.6 N for natural ones, and natural almonds broke into fewer, larger particles, "thus inhibiting the release of lipids".2 Their conclusion is one sentence and it is the one I would put on the fridge: "The amount of calories absorbed from almonds is dependent on the form in which they are consumed."2
A 2023 systematic review of 21 human and in vitro studies found the same ordering holds across nut types, with metabolisable energy rising from whole nuts to chopped to flour, and again from raw to roasted to nut butter.5
Read that table as a calorie tracker and the problem is obvious. The spread between whole almonds and almond butter is 59 kcal on the same 28 g. No food database has four entries for this, no barcode carries it, and a photograph cannot tell you whether the almonds in a salad arrived whole or chopped. It is the one variable here that actually moves the number, and it is the one nothing on your phone can see.
Whole grains, and 92 calories a day out the other end
Nuts are the best-measured case but they are not the only one. A controlled-feeding trial put 81 men and postmenopausal women, aged 40 to 65 with a BMI under 35, through a two-week run-in and then six weeks on one of two weight-maintenance diets. The diets differed in whole grains and fibre and were otherwise similar: 207 ± 39 g of whole grains and 40 ± 5 g of fibre a day on one arm, no whole grains and 21 ± 3 g of fibre on the other. Body weight was held steady by design.7
Three things moved on the whole grain arm. Stool weight rose by 76 g a day, stool energy content by 57 kcal a day, and resting metabolic rate by 43 kcal a day. Combined, the authors report "a 92-kcal/d (95% CI: 28, 156-kcal/d) higher net daily energy loss".7
Three caveats belong with that 92, and the authors supply two of them. When non-adherent participants were excluded, the between-group differences in resting metabolic rate and in prospective consumption were no longer statistically significant, which means the 43 kcal half of the total is the shakier half.7 The confidence interval runs from 28 to 156, so the honest reading is somewhere between trivial and worth noticing. And the fibre difference being tested was 19 g a day, which is a larger change than most people will make and sustain.
One more thing I would want told to me: one of the authors was affiliated with the Bell Institute of Health and Nutrition at General Mills.7 That does not make the stool calorimetry wrong. It is the sort of detail that belongs next to the number rather than in a footnote nobody opens.
Absorption is one of two ways processing changes what a food does. The other is how much of it people eat, which the post on ultra-processed food and what a tracker cannot see covers.
Cooking runs the error the other way
Everything above makes your log read high. Cooking does the opposite, and it is the reason this post is not an argument that calorie counts are all inflated.
Harvard researchers fed mice meat and starchy tubers, raw or cooked, pounded or whole, and measured what happened to body mass while controlling for food intake and activity. Cooking substantially increased the energy gained from meat, and beat non-thermal processing for both the meat and the tubers. Their own reading of what this implies is the sentence a calorie tracker should sit with: the results "illuminate a weakness in current food labeling practices, which systematically overestimate the caloric potential of poorly processed foods."8
A follow-up did the same for a fat-rich food. Cooking peanuts consistently increased the energy mice gained per calorie, while blending them had no detectable energetic benefit. Faecal fat fell when the peanuts were cooked, and microscopy showed the cell walls and the oleosin layer that normally shield the lipids from digestive lipases had been disrupted.9
These are mice. I could not find a human balance study of cooked against raw meat done to the standard of the almond work, and I am not going to convert a mouse result into a percentage for your dinner. Take the direction as established and the size as unmeasured in humans.
It is also worth saying plainly how this sits beside the other post on cooking, because at a glance they look like they disagree. Raw or cooked says cooking removes water rather than calories, and that is right: the energy in the food is conserved, and the per-100-gram figure only rises because the denominator shrank. This post is about a different quantity, which is how much of that energy makes it across your gut wall. A roasted chicken breast holds the same calories it went in with and hands you slightly more of them.
Why a packet is allowed to say 170
None of this is a scandal, and the regulation is worth reading because it is more permissive than most people assume. Under 21 CFR 101.9(c)(1)(i), caloric content "may be calculated by the following methods", and then five of them are listed.11
A manufacturer may use the Atwater specific factors from table 13 of USDA Handbook No. 74. Or the general factors of 4, 4 and 9. Or those general factors with non-digestible carbohydrates and sugar alcohols removed, in which case "a general factor of 2 calories per gram for soluble non-digestible carbohydrates shall be used". Or specific food factors that the FDA has approved elsewhere in the regulations "or by other means, as appropriate". Or bomb calorimetry data, "subtracting 1.25 calories per gram protein to correct for incomplete digestibility".11
Five legal methods, one food. Two companies packing the same almonds can declare different calorie counts and both be compliant, which is worth remembering the next time two apps disagree by fifteen calories and you assume one of them has a bad database.
Read the list again and notice what is missing. Not one of the five is the faecal balance method the nut studies used. The closest is bomb calorimetry, and its single correction is for protein digestibility, which is not where the almond gap lives. The measured almond value has been in print since 2012 and the packet in front of you still carries a calculated one.
That regulation has a second half that matters just as much and belongs to the other post: whichever of the five methods produced the declared number, the food may legally contain up to 20% more calories than it says. The size of that allowance, and the four other errors stacked around it, are covered in how accurate are calorie tracking apps.
What the evidence does not settle
The 2023 systematic review is the best summary of how far this goes, and it is also the best summary of how far it does not. Searching five databases to June 2021, it found 21 studies covering almonds, cashews, hazelnuts, pistachios, walnuts and peanuts. Metabolisable energy was "consistently lower than that predicted by Atwater factors" for every nut type investigated, with pistachios at 22.6 kJ/g and raw almonds at 18.5 kJ/g.5
Then the authors mark their own limits. The review "did not consider the digestibility of carbohydrates and protein, which should be considered when interpreting the outcomes", and the strength of the evidence base "was reduced by the variation in methods used between studies, suggesting that further clinical trials are needed to determine the impact of the findings of this review for clinical dietetics."5 That is a research field describing itself as promising and unfinished.
Four gaps I would want flagged if I were the one reading this.
- It covers six nut types. Not vegetables, not meat, not dairy, not bread, not a single mixed dish. For almost everything you will log this week, nobody has run the measurement and the calculated figure is all there is.
- The three headline studies enrolled 18 people each, ran at the same USDA centre in Beltsville and share authors. Consistency from one laboratory is not the same thing as replication across several.
- Individual variation is real and unexplained. In the cashew study the measured energy of one 28 g serving ranged from 105 to 151 kcal across participants.4 Whether that is your gut differing from mine or noise in the method, 18 people cannot tell you.
- Nobody has shown this changes anybody's weight. The review offers the lower energy as something that "may explain a lack of association between nut intake and body weight observed in the literature",5 and a candidate explanation for an epidemiological association is several steps short of an effect you could plan around.
What this means for calorie tracking, and what it does not
Start with the direction, because it is the part people get backwards. This error makes your log read too high. The dominant error in food logging runs the other way: measured against doubly labelled water, people under-report their own intake by roughly 15% on a 24-hour recall.14 If you eat whole nuts most days and misreport the way most people do, the two are pulling against each other and you have no way of knowing the net. Correcting the one you can see does not move you toward the truth, it moves you away from it by an amount you cannot check.
So do not subtract anything from your log. Four reasons, in the order I find them convincing.
The correction exists for six nuts. Apply it there and nowhere else and you have not made your log more accurate, you have made it inconsistently wrong, which is worse: a bias that applies evenly to every day mostly cancels when you compare Tuesday with Wednesday, and a bias that applies to Tuesday's almonds only does not.
The sizes are small. Forty calories on a 28 g serving of almonds, against a typical deficit of 500 a day, is 8% of the thing you are trying to measure and comfortably inside the error of everything else in the log. You would need to eat nuts by the bagful for this to outrun the noise, and at that point the portion estimate is the bigger problem.
No tracker can do it for you, including mine. MyPlate has no food database and no barcode scanner, a scan is an estimate rather than a measurement, and nothing in a photograph tells the model whether those almonds were chopped. I would rather say that than imply the app is applying a metabolisable energy correction behind the scenes, because it is not, and neither is anything else on the App Store.
Your scale already contains the answer. Whatever fraction of your almonds you absorb, a weight trend averaged over two to four weeks has it baked in. That is what the log is for: you adjust the target you are aiming at until the trend does what you want, and the absolute truth of any single number stops mattering.
One thing here is worth acting on, and it costs nothing. If nuts are already part of your week, whole ones deliver measurably less energy than the same weight of nut butter, 4.42 against 6.53 kcal/g,2 and you get that without logging anything differently. This is not a reason to start eating nuts to lose weight, and it is not a licence to treat a handful as free.
Nothing on this page is a reason to lower your calorie target. The gap described here means your log reads higher than what you absorbed, so if it argues for anything it argues you have slightly more room than you thought, not less. Targets also have a floor for good reason: MyPlate will not show a daily figure below 1,200 kcal for women or 1,500 for men, a limit taken from the NIH clinical guidelines on obesity.13 If reading about measurement error leaves you wanting to eat less to be safe, that is the instinct to be careful of, not the arithmetic.
And if tracking has started to feel like something being done to you rather than something you chose, put it down and talk to a doctor or a registered dietitian. No amount of precision about almonds is worth that.
Common questions
Do nuts really have fewer calories than the label says?
Yes, for the six nut types anyone has measured. A USDA balance study fed 18 adults fully controlled diets containing almonds and collected all urine and faeces for nine days: a 28 gram serving delivered 129 kcal against the 168 to 170 the Atwater calculation predicts, which the authors call a 32% overestimation.1 Walnuts measured 21% below the predicted value3 and cashews 16% below typical labels.4 The reason is physical: fat held inside intact plant cell walls is never fully released to digestion. It has only been measured for almonds, cashews, hazelnuts, pistachios, walnuts and peanuts, in studies of roughly 18 people each, so treat it as established for whole nuts and unmeasured for everything else on your plate.5
Should I log fewer calories for almonds than my calorie counter app says?
No. Correcting one food and nothing else makes a log inconsistently wrong rather than more accurate, and a bias that repeats evenly across days mostly cancels when you compare one day against another. The size is small in practice, around 40 kcal on a 28 gram serving,1 which sits inside the error of guessing the portion. Your own reporting error also runs the opposite way, since people under-report their intake by roughly 15% when it is checked against doubly labelled water,14 so fixing the smaller error alone moves you further from the truth rather than closer. Log the standard figure every day and let the weight trend over two to four weeks tell you whether the target needs moving.
Does cooking add calories to food?
Not to the food, but it does raise how many of them you absorb. The energy in a chicken breast is unchanged by roasting, and the per-100-gram figure only rises because water left. What changes is how much of that energy survives the trip through your gut. In mice, cooking substantially increased the energy gained from meat and from starchy tubers and beat non-thermal processing at both,8 and cooking peanuts increased energy gained per calorie while blending them did not, with faecal fat falling and the cell walls shielding the lipids visibly disrupted.9 Those authors wrote that current food labelling systematically overestimates the caloric potential of poorly processed foods.8 No human balance study of cooked against raw meat has been run to the same standard, so the direction is established and the size is not.
Why do two apps give different calories for the same packaged food?
Partly because US labelling permits five different calculation methods and the manufacturer picks one. Under 21 CFR 101.9(c)(1)(i) the calories in a food may be figured from the Atwater specific factors in USDA Handbook No. 74, from the general factors of 4, 4 and 9, from those general factors with non-digestible carbohydrates removed and soluble ones counted at 2 calories per gram, from specific food factors the FDA has approved elsewhere, or from bomb calorimetry with 1.25 calories per gram subtracted from protein for incomplete digestibility.11 Two companies packing the same product can declare different numbers and both be compliant. The rest of the gap comes from which database entry each app happened to pick.
Is a calorie and nutrient tracker still worth using if the numbers are conventions?
Yes, provided you expect it to measure change rather than truth. A convention applied the same way on Monday and on Thursday still reports which day was bigger, and that comparison is what any decision rests on. What it cannot hand you is a true figure for one meal, and no app on the market can: no consumer tracker applies measured metabolisable energy, because those values exist for six nut types and nothing else.5 So judge a tracker on whether you will still be opening it in a month, rather than on which one advertises itself as the most accurate calorie tracker.
Log the convention, watch the scale
The number in your food log is a set of averages from the 1890s, refined once in 1955, and for the three nuts measured most closely the energy comes in 16% to 24% below the calculated figure.134 That is a reason to be clear about what kind of instrument you are holding, rather than a reason to distrust calorie tracking.
Difference is what it measures well. The same convention applied to Monday and Thursday still reports which was bigger, and that is the comparison a decision actually rests on. It will not hand you the truth about a meal, and the Atwater problem is one more reason why, sitting alongside the label tolerance, the database entry and your own reporting.
So log the almonds at whatever number your app gives you, log them the same way tomorrow, and let two to four weeks of weight decide whether the target is right. The one piece of this you can act on does not involve the app at all: eat the nut rather than the butter, and change nothing in your log.2
Sources
Every figure above traces to one of these. If you find a number that doesn't match the source it claims, tell me and I'll correct it.
- Novotny JA, Gebauer SK, Baer DJ (2012). Discrepancy between the Atwater factor predicted and empirically measured energy values of almonds in human diets. American Journal of Clinical Nutrition, 96(2):296-301. Read on PubMed →
- Gebauer SK, Novotny JA, Bornhorst GM, Baer DJ (2016). Food processing and structure impact the metabolizable energy of almonds. Food & Function, 7(10):4231-4238. Read on PubMed →
- Baer DJ, Gebauer SK, Novotny JA (2016). Walnuts consumed by healthy adults provide less available energy than predicted by the Atwater factors. Journal of Nutrition, 146(1):9-13. Read on PubMed →
- Baer DJ, Novotny JA (2018). Metabolizable energy from cashew nuts is less than that predicted by Atwater factors. Nutrients, 11(1):33. Read on PubMed →
- Nikodijevic CJ, Probst YC, Tan SY, Neale EP (2023). The metabolizable energy and lipid bioaccessibility of tree nuts and peanuts: a systematic review with narrative synthesis of human and in vitro studies. Advances in Nutrition, 14(4):796-818. Read on PubMed →
- Traoret CJ, Lokko P, Cruz AC, et al. (2008). Peanut digestion and energy balance. International Journal of Obesity, 32(2):322-328. Read on PubMed →
- Karl JP, Meydani M, Barnett JB, et al. (2017). Substituting whole grains for refined grains in a 6-wk randomized trial favorably affects energy-balance metrics in healthy men and postmenopausal women. American Journal of Clinical Nutrition, 105(3):589-599. Read on PubMed →
- Carmody RN, Weintraub GS, Wrangham RW (2011). Energetic consequences of thermal and nonthermal food processing. Proceedings of the National Academy of Sciences, 108(48):19199-19203. Read on PubMed →
- Groopman EE, Carmody RN, Wrangham RW (2015). Cooking increases net energy gain from a lipid-rich food. American Journal of Physical Anthropology, 156(1):11-18. Read on PubMed →
- FAO (2003). Food energy: methods of analysis and conversion factors. FAO Food and Nutrition Paper, 77. Chapter 3, Calculation of the energy content of foods. Read chapter 3 →
- US Food and Drug Administration (2026). Nutrition labeling of food, 21 CFR 101.9(c)(1)(i). Code of Federal Regulations, Title 21. Read the regulation →
- US Department of Agriculture, Agricultural Research Service. FoodData Central, SR Legacy entry for nuts, almonds (FDC ID 170567). Search FoodData Central →
- National Heart, Lung, and Blood Institute, NIH (1998). Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. NIH Publication 98-4083. Read the guidelines →
- Freedman LS, Commins JM, Moler JE, et al. (2014). Pooled results from 5 validation studies of dietary self-report instruments using recovery biomarkers for energy and protein intake. American Journal of Epidemiology, 180(2):172-188. Read on PubMed →
The formulas and limits behind MyPlate's own numbers, with their citations, are on the health sources page.