Body fat percentage — methods, accuracy, and what each device really measures

The human body is modelled as compartments. The simplest is **two-compartment (fat mass + fat-free mass)**, first proposed by Behnke in 1942 using density measurement. Fat has density ~0.9 g/cm³; fat-free mass ~1.1 g/cm³. Siri's 1961 equation converts density to fat percentage: **%BF = (495/ρ − 450)**. Modern reference methods (four-compartment models) add water, mineral, and protein separately — the gold standard against which everything else is validated.

The methods, ranked by accuracy

| Method | SEE vs 4-C model | What it measures | Cost/session | |---|---|---|---| | 4-compartment model (research) | Reference | Body density + total body water + bone mineral | $500+ | | **DEXA** (dual-energy X-ray) | ±1.0–1.8% | X-ray attenuation at two energies → fat, lean, bone | $60–150 | | **Hydrostatic weighing** | ±2.0–2.5% | Whole-body density via underwater weight | $75–150 | | **BodPod** (air displacement) | ±2.0–2.7% | Body density via air-displacement plethysmography | $50–100 | | **3-site skinfold** (Jackson-Pollock) | ±3.0–4.0% | Subcutaneous fat thickness at 3 sites | $20–50 | | **7-site skinfold** (Jackson-Pollock) | ±2.7–3.5% | 7 sites, better whole-body sampling | $25–75 | | **US Navy tape** (Hodgdon-Beckett) | ±3.0–5.0% | Neck/waist(/hip) circumference regression | Free | | **Multi-frequency BIA** (InBody 570, Seca mBCA) | ±3.0–5.0% | Bioelectrical impedance at multiple frequencies | $40–75 | | **Consumer scale BIA** (Withings, Renpho) | ±4.0–8.0% | Impedance foot-to-foot only | $50–200 device | | **3D body scanner** (Naked, Styku) | ±3.0–5.0% | Circumferences + volume → density model | $150–300 device | | **Photo AI** (apps like MeThreeSixty) | ±4.0–7.0% | Silhouette from 4 photos + regression | Free–$20 |

SEE = standard error of estimate vs the 4-compartment reference in validation studies (Ackland 2012, Wang 1998). A method with ±3% SEE means a true 18% body fat person will typically read between 15% and 21% on any given measurement.

DEXA — the practical gold standard

Dual-energy X-ray absorptiometry uses two X-ray beams (40 and 70 keV). Fat, lean tissue, and bone attenuate the two energies at different ratios, allowing three-compartment separation across each pixel. A whole-body scan takes 5–10 min, dose ~0.5 μSv (less than 2 hours of natural background). **Manufacturer differences matter**: a Hologic scanner reads ~1–2% lower fat than a GE Lunar for the same person, so track trends on the same machine only. Hydration state shifts DEXA fat readings by 0.5–1% per litre of body water fluctuation — schedule morning scans, fasted, before training, on the same weekly day.

BIA — why it's so variable

Bioelectrical impedance passes a 500 μA current at 50 kHz (single-frequency) or multiple frequencies through the body. Fat is a poor conductor; lean mass (73% water) is a good conductor. The device measures impedance (Z) and applies a regression equation calibrated on a specific population. The reason a home scale reads 15% before coffee and 22% after a workout is **not fat gain in 90 minutes** — it's hydration:

+1 L body water → BIA fat reading typically −2 to −4% > Meal in stomach: +1–2% (extra current path) > Recent exercise (skin blood flow up): −2 to −3% (until cooled) > Cold hands/feet (peripheral vasoconstriction): +2 to −4% (varies) > Contact resistance dry skin: +5 to +10% wild readings

**BIA protocol for consistency:** morning, post-void, no exercise the prior 12 hr, no food/drink 4 hr, no alcohol 24 hr, warm room, dry electrodes. Follow this and even a consumer scale tracks *trends* within ±1.5% run-to-run.

Skinfold — Jackson-Pollock 3-site

Skinfolds measure subcutaneous fat with a spring-loaded caliper (Harpenden, Lange, or Slim Guide) at anatomical landmarks. Body density is estimated from the sum of skinfolds, then Siri's equation converts to %BF.

**Men (Jackson-Pollock 3):** chest, abdomen, thigh > **Women (Jackson-Pollock 3):** triceps, suprailiac, thigh > Sum-of-3 (mm) → density → %BF via age-adjusted regression > Same technician, morning of, hydrated, no exercise 24 hr. Different technicians disagree by 20–30% on the sum.

Skinfolds fail for **visceral-heavy body types** — a man with 4% subcutaneous fat but 15% visceral fat (thin outside, fatty liver) will read low. DEXA and MRI expose this; calipers cannot.

US Navy tape — the free method

Developed by Hodgdon & Beckett (Naval Health Research Center, 1984) as a screening tool. Only requires a tape measure:

**Men:** %BF = 86.010 × log10(waist − neck) − 70.041 × log10(height) + 36.76 > **Women:** %BF = 163.205 × log10(waist + hip − neck) − 97.684 × log10(height) − 78.387 > All measurements in inches. Waist at the navel, neck below Adam's apple, hip at widest point.

SEE ≈ 3.5% for the average recruit; worse for lean athletes (over-reports fat) or very obese (under-reports). Simple, requires no equipment, tracks trend well.

Reference ranges (ACE, ACSM 2022)

| Category | Men (%BF) | Women (%BF) | |---|---|---| | Essential fat | 2–5 | 10–13 | | Athletes | 6–13 | 14–20 | | Fitness | 14–17 | 21–24 | | Acceptable | 18–24 | 25–31 | | Obese | ≥25 | ≥32 |

Essential fat (organ padding, cell membranes, sex-specific stores) cannot be lost without health consequences. Bodybuilders on stage at 3–4% are in a temporary catabolic state, not sustained.

Worked example — reconciling three readings

35-year-old male, 80 kg, 178 cm, casual gym-goer. Same morning:

Home BIA scale (Withings): **21.4%** > Navy tape (34" waist, 15" neck): 86.010 × log10(34−15) − 70.041 × log10(70.08) + 36.76 = 86.010 × 1.279 − 70.041 × 1.845 + 36.76 = **19.7%** > 3-site skinfold (sum 42 mm): **17.8%** > DEXA (baseline, Hologic): **19.1%**

All within their expected SEE of the DEXA reference. Track *change* using one method with strict protocol — never mix methods month to month. A 2% drop on the same DEXA scanner over 3 months is real; a 2% drop switching from calipers to BIA is measurement noise.

Visceral fat — what none of the above measure well

Visceral adipose tissue (VAT) around abdominal organs is the metabolically dangerous fat. **DEXA android/gynoid ratio** correlates ~0.8 with MRI-measured VAT; consumer methods correlate 0.4–0.6. The **waist-to-height ratio** (should be < 0.5) is a simpler VAT proxy than any %BF reading — a 40-year-old at 22% BF with waist/height = 0.58 is at higher metabolic risk than a 25% BF person with waist/height = 0.48.

The five-second recap

DEXA is practical gold standard. BIA is noisy — protocol matters more than device. Skinfolds are accurate with the same technician. Navy tape is free and surprisingly close. Never mix methods; track trend on one, not absolute number.