Abstract
Background: Converting skinfold thicknesses into body fat percentage (%BF) assumes that subcutaneous adipose tissue (SAT) represents a fixed fraction of total fat, an assumption untested at extreme leanness. This study assessed whether skinfold–dual-energy X-ray absorptiometry (DXA) disagreement is random or structured by regional SAT. Methods: Thirty-five elite (Tier 4) female trail runners underwent anthropometry (Withers equation) and DXA in one session. Agreement was evaluated via Bland–Altman analysis and Lin's concordance correlation coefficient (CCC). The inter-method difference (%BF_GAP = %BF_WITHERS − %BF_DXA) was regressed on android SAT, adjusted for total adiposity, and benchmarked against DXA-derived negative controls. Results: The methods correlated closely (r = 0.866, 95% CI: 0.749 to 0.931) but showed poor concordance (CCC = 0.311). Skinfolds systematically underestimated DXA by 5.50 percentage points (pp; total error 5.74 pp), displaying proportional bias that widened with adiposity. Android SAT pre-dicted %BF_GAP (R² = 0.489). Residualized on total fat mass, SAT retained an independent contribution (ΔR² = 0.106), which no negative control showed. Propagated skinfold imprecision (SD = 0.19 pp) accounted for only ~1.3% of the observed difference variance (SD = 1.66 pp). Conclusions: The disagreement is systematic, phenotype-dependent, and structured by regional SAT, reflecting accuracy rather than precision limitations. Con-verted skinfold %BF should not be reported alone or compared against DXA reference data; raw skinfold sums are preferable, as a variable to be observed rather than a target to be reduced.