Body Composition Testing: DEXA vs BIA Accuracy

Step on a scale that promises your body fat to two decimal places, 22.37 percent, and the number feels like a fact handed down from a machine that has looked inside you. It hasn't. According to Grant Tinsley, professor of human performance at Texas Tech University and director of its Energy Balance and Body Composition Laboratory, no existing technology actually measures fat mass, muscle mass, or any other body composition value in a living person. Every device estimates it.

"Our ability to measure them is essentially zero, to measure them. Our ability to estimate them is really what we're getting at here with these different devices." (Grant Tinsley, [12:49])

That distinction sounds pedantic until you see what it explains. Tinsley spent a year dissecting cadavers during his master's training, the only true way to know, down to the gram, how much fat, water, protein, and mineral a body contains (04:02). Every other method, from a $20 bathroom scale to a $30,000 research-grade bioimpedance analyzer, is trying to infer that same breakdown from something indirect: how an electrical current passes through tissue, how X-rays are absorbed, how much a body displaces when submerged in water (15:56, 17:03).

Tinsley organizes these methods by what he calls levels. The molecular level treats the body as what he describes as "molecular soup" (09:25): fat is simply non-polar lipid, and fat-free mass is everything else, water, protein, mineral, and glycogen, wherever in the body it happens to sit (10:05). Most familiar tools, including DEXA (dual-energy X-ray absorptiometry, a scan that passes two X-ray energies through the body) and bioimpedance scales (devices that estimate water and fat content by measuring how easily electrical current passes through the body), operate at this level (17:21, 15:56). The organ-tissue level is different. It respects anatomy the way a cadaver dissection does, isolating actual muscle groups and fat depots. Skeletal muscle mass, the figure athletes and lifters care about most, lives here, and only imaging tools like MRI, CT, and ultrasound can really get at it (11:17, 18:16). The two levels produce numbers that sound related but are not interchangeable, which is why a DEXA printout and an MRI-derived muscle mass figure can disagree without either being wrong.

A Good Average Can Still Fail You

The more unsettling idea in the conversation is that a device can be statistically valid and still be wrong for any one person using it. Tinsley walks through a hypothetical: a bioimpedance scale is tested against DEXA in 100 people, the group averages land within 0.2 percentage points of each other, the correlation is a strong 0.97, and the statistics all say the methods agree (25:22). Good news, on paper. But plot every individual point, he says, and the picture can turn messy: some people overestimated by 10 percent, others underestimated by the same margin, canceling out in the average while leaving no individual confident in their own reading (26:07).

"You could certainly have a device that looks or that is valid at the group level, but for you as an individual, the hard thing is you don't know what data point you are." (Grant Tinsley, [27:17])

This matters most for people at the edges of the population used to build these equations: the very lean, the very muscular, people with unusually high or low body fat. Devices calibrated on an average body tend to perform worse the further someone is from that average (29:13), and most of the drift traces back to assumptions about fat-free mass itself, specifically how much of it is water. Cadaver research suggests fat-free mass is roughly 73 percent water on average, but that figure shifts across athlete populations and life stages, and when it shifts, every method built on that assumption inherits the error (30:45).

Some of that error is biological noise rather than a flaw in the device. Water, glycogen, and blood flow all move throughout a single day, so Tinsley's lab tests people after an overnight fast of 8 to 12 hours from both food and fluid, with deliberate hydration the night before (32:41). They also ask for roughly 36 hours of abstention from vigorous exercise, because something as ordinary as a leg workout changes local blood flow enough to shift a bioimpedance reading taken shortly after (33:28). When a full fast or rest period isn't realistic, in a clinic seeing patients in the afternoon, say, Tinsley's advice shifts from perfection to consistency.

"Standardize what you do at every assessment. The worst case scenario is you provide no input." (Grant Tinsley, [34:39])

That consistency matters most when someone is trying to detect a real change over weeks or months. Tinsley recommends a simple exercise: test a small group of people twice, on separate days, following identical pre-test procedures, and see how much the numbers move even though nothing biologically changed. If the largest swing observed is 2 percent, then any change smaller than that, after eight weeks of training, say, should be read as noise, not progress or regression (38:24). A 3 percent change, by contrast, starts to look like a real signal. The threshold is not universal; it depends on the specific device, model, and facility, which is why Tinsley resists giving one number that applies everywhere (39:52).

What the Scale at Home Is Actually Doing

Consumer devices complicate this further. Tinsley's lab has tested 14 different at-home bioimpedance scales, ranging from about $20 to $300, and found results all over the map (46:41). Some performed worse than having no data at all. Others approached or even exceeded laboratory-grade accuracy, with no obvious pattern, cheap versus expensive, foot-only versus foot-and-hand electrodes, predicting which was which (47:03). A follow-up trial in roughly 200 people, focused on more established manufacturers, found the same spread (47:37).

"There were some that were predictably terrible, like worse than not having the information. There were some that were really good, approaching or even exceeding laboratory grade." (Grant Tinsley, [47:03])

Tinsley's lab also noticed something stranger: some consumer scales appear to smooth or lock results across repeated uses, possibly to stop the number from bouncing around and frustrating the owner, which is good for peace of mind but bad for anyone hoping each reading is an independent measurement (49:30). His practical rule for anyone tracking body composition at home is to never compare numbers across two different devices or technologies; switching from a DEXA scan to a home scale should reset the baseline, not continue a trend line (57:13).

Underneath the technical detail is a quieter argument about what any of this data is for. Tinsley is careful to say that for some people, especially those already attuned to how their training and nutrition are going, a body composition number adds little beyond what they already know from how clothes fit or how a lift feels (58:21). For others, particularly people undergoing significant weight change, the composition of that change, how much is fat versus fat-free mass, can flag a real problem worth addressing (52:40).

"If the measurement produces anxiety or isn't informative, doesn't seem helpful, or doesn't add any information you don't already have, then it's certainly not necessary." (Grant Tinsley, [59:46])

Asked for one piece of advice to close the conversation, Tinsley didn't return to electrodes or X-rays. He suggested, instead, making room to think about mortality, about how small a single life is against the scale of the universe, as a way of clarifying what actually deserves attention (60:50). It is a strange pivot after an hour of phase angles and fasting protocols, but it lands on the same point the whole conversation was making. A number on a scale, however precise it looks, is never the whole picture. What a person does with the number, and how much weight they give it, is the part that was always up to them.