It has been consistently demonstrated that human proximal limb elements exhibit negative allometry, while distal elements scale with positive allometry. Such scaling implies that longer limbs will have higher intralimb indices, a phenomenon not borne out by empirical analyses. This, therefore, creates a paradox within the limb allometry literature. This study shows that these apparently conflicting results are the product of two separate phenomena. First, the use of the geometric mean of limb elements produces allometry coefficients that are not independent, and that when using ordinary least squares regression must yield an average slope of one. This phenomenon argues against using the geometric mean as a size variable when examining limb allometry. While the employment of relevant dimensions independent of those under analysis to calculate the geometric mean—as suggested by Coleman (Am J Phys Anthropol 135 (2008) 404–415)—may be a partial method for resolving the problem, an empirically determined, independent and biologically relevant size variable is advocated. If stature is used instead of the geometric mean as an independent size variable, all major limb elements scale with positive allometry. Second, while limb allometry coefficients do indicate differential allometry in limb elements, and thus should lead to some intralimb index allometry, this pattern appears to be attenuated by other sources of limb element length variation. Am J Phys Anthropol, 2011. © 2010 Wiley-Liss, Inc.