The effects that haemoglobin-metabolite interactions have on estimates of free magnesium-ion concentration in human erythrocytes, determined by 31P NMR [Gupta, R. K. et al., J. Biol. Chem. 253, 6172-6176 (1978)], were investigated. If the metabolite-haemoglobin association constants of Berger et al. [Eur. J. Biochem. 38, 553-562 (1973)] are used in the analysis then the estimates of intracellular free magnesium-ion concentration made by Gupta et al. (0.25 and 0.67 mM) become 0.43 and 0.60 mM, for oxygenated and deoxygenated cells, respectively. In oxygenated cells, this difference is primarily due to the lower value of KHbMgATP, given by Berger et al. These newly calculated concentrations are in closer agreement with those of Flatman (0.40 mM for oxygenated cells; 0.62 mM for deoxygenated cells) [Flatman, P. W., J. Physiol. 300, 19-30 (1980)] obtained with the 'zero-point titration' method. In addition, the assumptions that the chemical shift separations between the alpha- and beta-phosphorus resonances of ATP and MgATP are unchanged on association with Hb were shown to be false. Under normal intracellular conditions this may lead to errors of 5-10%. Much larger errors would be possible in cases where significant amounts of ATP or MgATP are bound to Hb. These outcomes place doubt on measurements of intracellular free Mg2+ concentration made using 31P NMR if there is no consideration given to the total concentration of 2,3-bisphosphoglycerate (BPG), ATP and Hb in the sample; the same principle would apply to other cell-types.