Flow-mediated dilation (FMD) is a noninvasive indicator of endothelial function and is routinely expressed as the percentage change in arterial diameter (FMD%) from a resting baseline (Dbase) to a postischemic peak (Dpeak). This expression is equivalent to the ratio of Dpeak/Dbase and is, therefore, dependent on important statistical assumptions, which have never been analysed in the context of FMD%. We aimed to investigate these assumptions, via a comparison of FMD between samples of children and adults, as well as to explore other approaches to scaling diameter change for Dbase. We found that FMD% did not scale accurately for interindividual differences in Dbase but, as expected, overestimated endothelial function for low Dbase and vice versa. We argue that this imprecise scaling of FMD% is predictable, not explained by physiology and is probably common. This problem is resolved by applying scaling principles, whereby the difference in diameter is the outcome and Dbase is a covariate in a logarithmic-linked generalized linear model. A specific allometric expression of FMD can be derived and we found this to be Dpeak/Dbase0.89 rather than a simple ratio in our particular dataset. We found that sample differences in endothelial function were inaccurate with FMD% versus our new allometric approach, and that FMD% misclassified participants into ‘high’ and ‘low’cohorts, which has implications for prognostic-type studies. We conclude that the general use of FMD% could have led to biased comparisons of different conditions and/or populations in past studies. Our new approach to scaling FMD is flexible for different datasets and is not based on the current assumption that a percentage change is appropriate in all circumstances.