Inside the Calorie Math: A Deep Dive Into How TDEE Formulas Are Built
Every calorie-counting app you've ever used is built on top of a small handful of regression equations derived from metabolic studies conducted decades ago — some of them before World War II. That's worth sitting with for a moment. The "personalized" number your fitness tracker spits out is the output of a formula fitted to a population of subjects who were probably nothing like you.
That's not a dismissal of these tools. It's an invitation to understand them well enough to use them intelligently. So let's crack open the math.
What TDEE Actually Measures — And What It Doesn't
Total Daily Energy Expenditure is the sum of four distinct components: Basal Metabolic Rate (BMR), the Thermic Effect of Food (TEF), Non-Exercise Activity Thermogenesis (NEAT), and Exercise Activity Thermogenesis (EAT). Of these, only BMR is calculated from a formula. The others are either estimated with an activity multiplier or ignored entirely.
BMR itself refers to the calories your body burns at complete physiological rest — not just sitting still, but lying motionless in a thermoneutral environment after a 12-hour fast. True BMR measurement requires indirect calorimetry: a metabolic cart that measures oxygen consumption and carbon dioxide production. The formulas we use are approximations of this, trained on measured data.
That distinction matters. When people say "the formula was wrong for me," they're often correct, but they're describing something that was never designed to be exact for any individual — it was designed to minimize average error across a population.
Harris-Benedict (1919, Revised 1984): The Original, Imperfect Blueprint
Francis Benedict and James Arthur Harris published their landmark equations in 1919, built from measurements on 239 subjects. The original equations are:
Men: BMR = 66.5 + (13.75 × weight in kg) + (5.003 × height in cm) − (6.775 × age in years)
Women: BMR = 655.1 + (9.563 × weight in kg) + (1.850 × height in cm) − (4.676 × age in years)
For 1919, this was genuinely impressive science. The problem is that the subjects were almost entirely young, healthy, white Americans — not a representative sample of modern populations. Body composition was never directly measured. The equations treat two people of identical height, weight, and age as metabolically identical regardless of how much of that weight is muscle versus fat.
Roza and Shizgal revised the equations in 1984 using a larger dataset (337 subjects), producing slightly different coefficients that perform marginally better. Most software today uses the revised version, though it's often still labeled simply "Harris-Benedict" without clarifying which revision.
The 1984 revision is more accurate on average, but it still carries the fundamental limitation: it encodes body weight, not body composition. A 90kg powerlifter and a 90kg sedentary person with significantly different fat percentages will get identical BMR estimates. Metabolically, they are not identical people.
Mifflin-St Jeor (1990): The Current Clinical Standard
Mark Mifflin and Sachiko St Jeor published a recalibration in 1990 using 498 subjects — a larger, more diverse sample that included obese individuals. Their equations:
Men: BMR = (10 × weight in kg) + (6.25 × height in cm) − (5 × age in years) + 5
Women: BMR = (10 × weight in kg) + (6.25 × height in cm) − (5 × age in years) − 161
The American Dietetic Association endorsed Mifflin-St Jeor as the most accurate single-equation predictor for non-athletes, and subsequent validation studies have generally confirmed this. A 2005 meta-analysis in the Journal of the American Dietetic Association found Mifflin-St Jeor to be accurate within 10% for roughly 82% of subjects — compared to about 70% for the revised Harris-Benedict.
The simplification in the Mifflin formula is somewhat counterintuitive: it uses cleaner coefficients (10, 6.25, 5 rather than the irregular numbers in Harris-Benedict), which suggests the 1984 revision may have been overfitted to its specific dataset. Mifflin-St Jeor's cleaner structure generalizes better.
Where it still struggles: lean athletes, elderly populations, and very obese individuals. For someone with unusually high or low lean body mass relative to their total weight, both Mifflin-St Jeor and Harris-Benedict will systematically over- or underestimate.
Katch-McArdle: The Lean Mass Equation
William Katch and Frank McArdle took a fundamentally different approach. Rather than regressing on total body weight, they used lean body mass as the primary predictor:
BMR = 370 + (21.6 × lean body mass in kg)
This is a more physiologically coherent model. Adipose tissue is metabolically inert compared to muscle, liver, and brain tissue. Fat mass contributes almost nothing to resting metabolism — somewhere around 4.5 calories per kilogram per day versus over 13 for muscle. Katch-McArdle accounts for this directly.
The equation also strips out height, age, and sex as separate variables. The reasoning is that once you control for lean mass, these variables mostly lose their predictive power. A 45-year-old woman and a 25-year-old man with the same lean body mass will have similar BMRs — and this turns out to be mostly true, particularly after adjusting for organ mass differences.
The catch: you need an accurate body fat percentage measurement to use this formula, and accurate measurement is harder than most people realize. Consumer DEXA scans carry error margins of around 1-2%. Bioelectrical impedance devices (the kind built into most bathroom scales) can be off by 5-8 percentage points under normal conditions, and much worse when hydration varies. Skinfold calipers are technique-dependent.
So Katch-McArdle is the most theoretically sound equation for people who actually know their body composition. It's the standard recommendation for bodybuilders, physique athletes, and anyone who has gotten a reasonably accurate DEXA scan. For someone relying on their scale's impedance reading after a salty dinner, it can be less accurate than Mifflin-St Jeor.
The Activity Multiplier Problem
Here's where TDEE calculations typically introduce more error than any of the BMR equations themselves. The standard multipliers — sedentary (1.2), lightly active (1.375), moderately active (1.55), very active (1.725), extremely active (1.9) — were derived from doubly-labeled water studies published in the 1980s and 1990s.
These multipliers are wide categories applied to widely varying individuals. "Moderately active" covers a spectrum from someone who takes 20-minute walks three times a week to someone doing daily two-hour cycling sessions. The actual PAL (Physical Activity Level) multiplier for a sedentary office worker can range from 1.4 to 1.6 depending on their NEAT — how much they fidget, stand, gesture, and move incidentally throughout the day. NEAT variation between individuals can exceed 2,000 calories per day, which dwarfs any differences between the BMR equations.
This is the deepest limitation of TDEE formulas: they're precision instruments applied to an inherently imprecise input. The BMR equations may be accurate to within 5-10%, but the activity categorization can introduce 15-25% error on top of that.
So Which Equation Should You Trust?
The honest answer is that you should treat any TDEE estimate as a starting point, not a prescription. But here's how to choose:
If you don't know your body fat percentage: Use Mifflin-St Jeor. It's the clinical default for good reason — best validated, widest applicability, endorsed by major nutrition bodies.
If you have a reliable body fat measurement (DEXA, hydrostatic, or a well-performed skinfold assessment): Use Katch-McArdle. It will outperform the others for anyone who deviates significantly from average body composition — lean athletes especially.
If you're using Harris-Benedict: Switch to Mifflin-St Jeor. There's no scenario where Harris-Benedict is the better choice among these three.
More practically: track your actual intake and weight change over three to four weeks, then back-calculate your real TDEE from the data. If you ate an average of 2,400 calories per day and your weight was stable, your TDEE is approximately 2,400 calories. This empirical approach, done carefully, will outperform any formula.
The Deeper Point About Population Equations
These equations are tools born from averaging. They describe the center of a distribution, not its edges. Someone with hypothyroidism, someone who has undergone prolonged caloric restriction and experienced adaptive thermogenesis, someone with unusually high organ mass — all of these individuals will sit far from the population mean, and the equations will fail them.
Understanding what a formula was built to do — and what population it was built from — is the only way to use it well. TDEE calculators are best understood not as calorie oracles but as calibration devices: a place to start counting, with the expectation that reality will correct you soon enough.