I Got My Body Fat Tested Every Way- Which Is Most Accurate?

Mike explained that because waist‑to‑hip ratio omits any tissue‑specific information, it is effectively useless for cross‑sectional body fat estimation. However, changes in waist circumference strongly covary with changes in total adiposity for a given person. He therefore repurposed the metric as a longitudinal tracking tool. The 3‑week interval is frequent enough to catch trends before they compound, yet long enough to filter out daily fluctuations. The stop‑bulk rule of hitting the cap twice consecutively prevents false stops from transient bloating. He noted that this method emulates the logic used in some old‑school military and insurance screening protocols, but applied intelligently over time.

The Bod Pod uses air displacement plethysmography: it measures body volume and, combined with mass, computes whole‑body density. The two‑compartment model then estimates fat and lean mass. Mike emphasised that this approach is identical in principle to underwater weighing but far more convenient. He acknowledged that its error rate means the result should be taken as an approximation, not gospel. He still recommended it as a practical endpoint measure because it avoids the $100 price tag and X‑ray exposure of DEXA, and the cold, wet experience of underwater weighing. For most people who do not have extreme bone density, the Bod Pod will land within about 3 percentage points of true fat mass.

Mike’s DEXA controversy taught him that firmware can corrupt the output. He therefore recommended using DEXA purely as one data point among several. For fat percentage, DEXA is generally valid regardless of hydration—unlike lean mass, which varies with body water. So he specifically counselled that the % fat number from DEXA is the safer metric, while the lean body mass kilogram value should be cross‑checked with rep strength records. If strength is stable or climbing and the DEXA shows only small changes in apparent LBM, the dieter can be confident that muscle was preserved. He warned against scanning every 3 weeks because short‑term fluctuations are noise, and against changing hydration protocols between scans (e.g., being carb‑loaded on one scan and depleted on the next), as that can produce spurious 5–10‑pound “muscle” losses or gains.

Mike highlighted that the 7‑site formula has the largest body of validation literature, making it the most defensible choice. He noted that his own results varied substantially with site selection: 9‑site gave 5.9%, 7‑site 8.3%, 3‑site 4.13%. The 7‑site omitted his very lean calves and biceps, leading to a higher (and likely more accurate) percent fat estimate. However, he argued that the actual number matters less than the consistent trend; if the 7‑site sum of skinfolds is decreasing at a steady rate over 8–12 weeks, the diet is working. He warned strongly against having an untrained person perform the test because “you’re kind of just gambling,” and encouraged finding a professional certified in anthropometry. He also suggested that combining skinfold tracking with waist circumference and strength data provides a robust, inexpensive monitoring package.

Mike framed this as the key take‑home from his own experience. After obtaining readings from 3.2% to 32.8%, he recognised that the BIA scale and handheld were absurd, the density‑based methods were biased low due to his bones, and the skinfolds (7‑site), DEXA, and waist‑to‑hip range pointed toward roughly 7–8%. He explicitly advised against picking the most flattering number and instead looking for a tight grouping. He argued that if three or four fundamentally different technologies agree to within a couple of percentage points, the odds are good that the real value lies in that interval. He cautioned that this approach still doesn’t replace laboratory‑grade 4‑compartment models, but for practical purposes it is vastly superior to trusting a single DEXA or BIA reading.

Mike explained that both the Bod Pod and underwater weighing determine body composition exclusively from whole-body density. The underlying equations use fixed assumed densities for lean tissue and fat. Because his bones are “like off the charts dense […] five standard deviations above the average,” his whole-body density sits well above the population mean. As a consequence, the models back-calculate less fat mass to make the numbers work, producing the fantastical readings of 3.2% (underwater) and 4.4% (Bod Pod). He emphasised that this is not a failure of the technology per se, but an inherent limitation of density-only two-compartment models when an individual’s lean mass composition deviates significantly from the norm. He cautioned that people with very frail bones could see the opposite bias, and that even underwater weighing, the research gold standard, carries a 1–2% error rate from bone density assumptions alone.

Mike argued that the DEXA system, while using X‑ray absorptiometry to quantify fat, lean, and bone mass separately, ultimately relies on proprietary “classified firmware” to calculate percentages. When a different brand of scanner or a different software version was used, the results swung wildly from 14% to 6%. He emphasised that the 14% reading was clearly erroneous because his visual leanness, strength maintenance, and other tests contradicted it. However, rather than naively trusting the 6% from the other scanner, he recognised the same black-box risk could apply. He concluded that DEXA remains a useful tool but must be interpreted with scepticism; using a DEXA at a single point to “prove” a certain body fat is dangerous because one bad firmware iteration can produce a face-value number that is miles from reality.

Mike’s approach mirrors the scientific method: use convergent evidence rather than a single assay. In his case, the waist‑to‑hip ratio gave 10–14%, skinfolds (7‑site) 8.3%, DEXA 6%, and the density‑based methods 3.2–4.4% after correcting for bone density. He reasoned that 3.2% and 32.8% (BIA) were clear outliers, while 6–8.3% formed a plausible cluster. He advocated using the densest cluster (around 7–8% in his case) as the best guess, and warned that if a method consistently deviates from the cluster in future, it probably should not be trusted. He did not claim this is a formal validation, but a practical heuristic for individuals who cannot access research‑grade multicompartment gold standards.

Mike explained that because no actual tissue is measured, the waist‑to‑hip ratio cannot distinguish whether a wide hip is due to fat or skeletal structure; thus, its cross‑sectional body fat estimate is “almost completely useless.” However, because fat loss and gain manifest proportionately at the waist in most individuals, repeated waist measurements act as a sensitive, low‑cost marker of energy balance direction. He proposed a concrete protocol: define your leanest waist (e.g., 34 inches) and set a hard upper limit for a bulking phase (e.g., 38 inches). If your waist hits 38 inches on two consecutive measurements (every 3 weeks), immediately transition to maintenance and then a fat loss phase. This prevents the “too fat” realisation after gaining 42 inches. He stressed that the system works because you always compare to your own previous data, eliminating genetic skeleton variability.

Mike observed that the 7‑site formula, which is the most validated in the research literature, gave 8.3%—closest to his presumed true value—while the 3‑site (4.13%) and 9‑site (5.9%) were lower. The discrepancy arose because the 9‑site and 3‑site omitted, respectively, his very lean calves and biceps, inflating the lean estimate. He argued that as long as the same sites and same technician are used each time, the trend in skinfold sums over months is reliable and informative; the absolute percent is noise. He strongly advised against interpreting a single skinfold‑derived percent as gospel, but endorsed having a certified anthropometrist take measurements every 3–4 weeks during a diet or bulk to gauge whether fat is being lost or gained too rapidly.

Mike pointed to the ‘athlete mode’ toggle on BIA devices as a tell‑tale sign that the underlying algorithm is not robust. He argued that human physiques exist on a continuum; partitioning people into ‘athlete’ and ‘normal’ modes is a crude post‑hoc correction that is unlikely to capture the differences in muscle mass, bone density, and hydration typical of trained individuals. His BIA scale reading of 32.8% body fat, alongside visible quad veins, glute striations, and a DEXA of 6%, illustrates the magnitude of error. He advised that anyone outside the validation demographic (older, female, jacked, or lean) should completely ignore BIA‑derived body fat numbers, though he conceded that for the narrow demographic it might work “pretty well”.