In localized brain proton MR spectroscopy ((1)H-MRS), metabolites' levels are often expressed as ratios, rather than as absolute concentrations. Frequently, their denominator is the creatine [Cr], which level is explicitly assumed to be stable in normal as well as in many pathologic states. The rationale is that ratios self-correct for imager and localization method differences, gain instabilities, regional susceptibility variations and partial volume effects. The implicit assumption is that these benefits are worth their cost(w)-(w) propagation of the individual variation of each of the ratio's components. To test this hypothesis, absolute levels of N-acetylaspartate [NAA], choline [Cho] and [Cr] were quantified in various regions of the brains of 8 volunteers, using 3-dimensional (3D) (1)H-MRS at 1.5 T. The results show that in over 50% of approximately 2000 voxels examined, [NAA]/[Cr] and [Cho]/[Cr] exhibited higher coefficients of variations (CV) than [NAA] and [Cho] individually. Furthermore, in approximately 33% of these voxels, the ratios' CVs exceeded even the combined constituents' CVs. Consequently, basing metabolite quantification on ratios and assuming stable [Cr] introduces more variability into (1)H-MRS than it prevents. Therefore, its cost exceeds the benefit.