Mifflin, Harris-Benedict, Cunningham: Resting Metabolism Methodology

Resting metabolic rate is the single load-bearing number in diet planning. Three equations dominate clinical and recreational use, derived from three different cohorts at three different points in the 20th century. They produce systematically different predictions, and the choice between them affects the calorie-budget math meaningfully. This article walks the math, the calibration cohorts, and where each equation breaks.

The formulas

Harris-Benedict (1919, original):
  Male:   66.5 + 13.75×wt + 5.0×ht − 6.76×age
  Female: 655 + 9.56×wt + 1.85×ht − 4.68×age

Harris-Benedict (Roza & Shizgal 1984 revision):
  Male:   88.4 + 13.4×wt + 4.8×ht − 5.7×age
  Female: 447.6 + 9.25×wt + 3.1×ht − 4.33×age

Mifflin-St Jeor (1990):
  10×wt + 6.25×ht − 5×age + 5 (male) / −161 (female)

Cunningham (revised 1991):
  500 + 22 × lean_body_mass_kg

All produce kcal/day. Weight in kg, height in cm, age in years, LBM in kg.

Derivation: how each equation emerged

Harris-Benedict (1919)

Harris and Benedict measured RMR in 239 subjects using indirect calorimetry — measuring oxygen consumption at rest, calibrated against carbon dioxide production. The cohort was American, predominantly white, and lived in the era before the modern adiposity profile became typical. The 1919 cohort's average body composition was leaner than the modern population's, which is why the unmodified equation systematically over-predicts modern adults' RMR by 5–10%.^[2]

The Roza & Shizgal 1984 revision recalibrated the constants against a more recent cohort but kept the same equation structure. The revised version still over-predicts modern adults, particularly those with higher body fat, because the underlying assumption (linear weight-height-age relationship) no longer matches population body composition.^[2]

Mifflin-St Jeor (1990)

Mifflin and colleagues recalibrated the equation on 498 healthy adults across a wider age range and BMI distribution than Harris-Benedict's cohort. The simpler integer-friendly coefficients (10, 6.25, 5) were chosen to produce comparable accuracy with simpler mental math. The 1990 cohort more closely represents modern adult populations, which is why Mifflin has displaced Harris-Benedict as the population default in clinical and recreational use.^[1]

The female sex constant (−161) was particularly carefully calibrated; earlier equations under-estimated female RMR by 8–12% because the cohorts were under-sampled.

Cunningham (1980/1991)

Cunningham took a structurally different approach: instead of predicting RMR from anthropometric inputs (weight, height, age), the equation predicts directly from lean body mass. The intuition is that RMR scales primarily with lean tissue's metabolic activity, with fat mass contributing a smaller and more variable share. For subjects with measured LBM (DXA, hydrostatic, BIA), Cunningham eliminates the cohort-mismatch problem because the predictor variable is body-composition-specific.^[3]

The trade-off is that the equation requires LBM as input. For population use without DXA access, Mifflin's anthropometric prediction is the only practical choice.

The empirical record

Three studies anchor the comparison:

1. Frankenfield et al. 2005 — compared Mifflin, Harris-Benedict, Owen, and Schofield against measured RMR in 408 healthy adults. Mifflin had the smallest mean error and the narrowest 95% CI; Harris-Benedict over-predicted by 5–8% across the cohort.^[2]
2. Cunningham 1991 revision — pooled data across multiple measured-RMR studies to derive the lean-mass-based equation. Found LBM explains roughly 80% of inter-individual RMR variance, with weight, height, and age adding only ~5% more.^[3]
3. Madden et al. 2013 systematic review — compared RMR equations specifically in obese vs lean adults. Found Mifflin under-predicts in obese subjects by 5–7% (RMR is higher than the equation predicts because the visceral organ mass scales with body size). Cunningham held closer accuracy across the BMI range because LBM is the primary predictor.^[4]

Worked predictions for an 80 kg male, 178 cm, age 30

Mifflin-St Jeor:    10×80 + 6.25×178 − 5×30 + 5  = 800 + 1112.5 − 150 + 5  = 1767.5 kcal
Harris-Benedict:    88.4 + 13.4×80 + 4.8×178 − 5.7×30  = 88.4 + 1072 + 854.4 − 171 = 1843.8 kcal
Cunningham (68 kg LBM):  500 + 22×68 = 500 + 1496 = 1996 kcal

Three equations, three different RMR estimates: 1767.5, 1843.8, 1996 kcal. The 76-kcal spread between Mifflin and Harris-Benedict is the published over-prediction; the 228-kcal spread to Cunningham reflects that the lifter has above-average lean mass (assumed 68 kg LBM at 15% body fat), which the anthropometric equations cannot see.

Where each equation fails

Harris-Benedict

Systematically over-predicts modern adults by 5–8%. The mechanism: the 1919 calibration cohort had lower body fat than today's average adult, so the equation's coefficient on weight is calibrated against a leaner population. Applied to today's adults, the equation interprets extra body mass as if it were extra lean mass, over-predicting metabolic activity.^[2]

Mifflin-St Jeor

Mifflin holds well for healthy adults across a wide BMI range but under-predicts in two specific populations: obese subjects (visceral organ mass elevates RMR above the anthropometric prediction) and trained athletes (lean-mass excess elevates RMR above the anthropometric prediction). Both errors are in the same direction (under-prediction) but driven by different mechanisms.^[4]

Cunningham

Cunningham is accurate when LBM is measured. The failure mode is when LBM is estimated rather than measured. Estimated LBM (from height-weight equations or BIA scales) carries ±3–5 kg error, which propagates linearly through Cunningham's 22-kcal-per-kg coefficient: a 5 kg LBM estimation error produces a 110 kcal RMR error, larger than Mifflin's typical error.^[3]

Bottom line: which equation for which case

1. Population default, no body-comp data: Mifflin-St Jeor. Best accuracy without LBM input.
2. Trained athlete with DXA-measured LBM: Cunningham. Captures the lean-mass excess Mifflin misses.
3. Obese subject with measured LBM: Cunningham. Better than Mifflin which under-predicts obese RMR.
4. Anyone using BIA-scale LBM: Mifflin. The BIA error compounds into Cunningham more than it adds value.
5. Harris-Benedict: Use only for historical comparison with older programs. The 1919 cohort no longer represents modern adults.

The TDEE pipeline

RMR is the input to TDEE: TDEE = RMR × activity factor. The standard activity factors (1.2 sedentary, 1.375 light, 1.55 moderate, 1.725 very active, 1.9 extreme) carry their own ±10% error band, which compounds with the RMR estimation error. For the worked 80 kg male example at moderate activity:

Mifflin × 1.55:        2739 kcal TDEE
Harris-Benedict × 1.55: 2858 kcal TDEE
Cunningham × 1.55:     3094 kcal TDEE

The 355-kcal spread between Mifflin and Cunningham at the TDEE stage exceeds the 228-kcal spread at RMR — the activity-factor multiplier amplifies the underlying RMR difference. For a trained lifter with measured LBM, the Cunningham-based TDEE is the appropriate anchor for diet planning.^[3]

Adaptive thermogenesis: where the equations get smaller

All three equations predict static RMR. In practice, RMR drops during sustained caloric deficits — the published adaptive-thermogenesis literature places the reduction at 5–15% below baseline after 8–12 weeks of moderate deficit, with full recovery taking 4–8 weeks after returning to maintenance. The equations don't see this drop because they were calibrated on weight-stable subjects.^[4]

Practical consequence for diet planning: a Mifflin-derived TDEE that worked at week 1 of a cut will systematically overestimate maintenance calories by week 8. The published correction is to re-measure RMR every 4–6 weeks during long cuts, or to adjust the activity factor downward by 0.05–0.10 to compensate. Lifters who follow Mifflin's static prediction across a 16-week cut typically find the cut stalls around week 10 — which is the adaptive-thermogenesis signal, not a tracking failure.

Cross-checking against related tools

The BMR Calculator exposes Mifflin and Harris-Benedict for direct comparison. The TDEE Calculator handles the activity-factor multiplier. The Calorie Deficit Calculator sits one step downstream, converting TDEE into cut targets.

Related reading: TDEE Formulas Compared for the cross-engine comparison, TDEE For Athletes for the activity-factor selection at high training volumes, and How To Break A Weight Loss Plateau for the adaptive-thermogenesis effects that erode the RMR-based budget over time.

FAQ

Why does Harris-Benedict still appear in textbooks?

Inertia and citation conservatism. Many clinical textbooks list both Harris-Benedict and Mifflin without preference, partly because the equations produce similar enough numbers in lean populations that the historical equation's continued mention is not strictly wrong. Modern practice has moved to Mifflin as the default; Harris-Benedict survives as historical reference.^[2]

Does Cunningham work without DXA?

Less well. The equation's accuracy comes from accurate LBM input. BIA-scale LBM estimates carry ±3–5 kg error, which propagates into Cunningham at 22 kcal/kg. The net result is that Cunningham with BIA-derived LBM is no more accurate than Mifflin with simple anthropometric inputs.^[3]

How much error should I expect in my TDEE estimate?

±10–15% from the combination of RMR estimation error (±5–8%) and activity-factor selection error (±10%). For a TDEE estimate near 2700 kcal, that is roughly ±300 kcal of plausible range. Adjust by tracking bodyweight at a fixed intake for 4 weeks rather than chasing the formula precision.^[4]

What about the Katch-McArdle equation?

Katch-McArdle is essentially Cunningham with a slightly different intercept (370 vs 500) and coefficient (21.6 vs 22). The two equations produce predictions within 1–2% of each other. Use either when LBM is measured; the choice between them rarely affects practical decisions.^[3]

References

1. 1 A new predictive equation for resting energy expenditure in healthy individuals (Mifflin-St Jeor) — American Journal of Clinical Nutrition (1990)
2. 2 Comparison of predictive equations for resting metabolic rate in healthy adults — Journal of the American Dietetic Association (2005)
3. 3 Body composition and resting metabolism: the Cunningham equation revisited — American Journal of Clinical Nutrition (1991)
4. 4 RMR equations in obese vs lean adults: a systematic review — Obesity (2013)