The total Mg2+ content of human red cells ([Mg]T,i) is partitioned between free and bound forms. The main cytoplasmic Mg2+ buffers are ATP and 2,3 bisphosphoglycerate. Haemoglobin binds free ATP and bisphosphoglycerate, preferentially in the deoxygenated state. Thus, the free ionized Mg2+ concentration ([Mg2+]i) oscillates with the oxy-deoxy condition of the cells. The binding reactions are also modulated by the pH changes that accompany the oxygenation-deoxygenation transitions. The complex interactions between Mg2+, its ligands and Hb can be encoded in a set of equilibrium equations representing all the known binding reactions of the system. To develop a comprehensive understanding of the Mg2+ homeostasis of intact red cells it is necessary to correct and refine the equations and parameters of the model by systematic comparisons between model predictions and measured cytoplasmic Mg2+ buffering curves under a variety of experimental conditions. Earlier models largely underestimated total Mg2+ binding in intact cells. We carried out experiments in which [Mg]T,i and [Mg2+]i were controlled over a wide range ([Mg]T,i between 0.1 and 23 mM) by the use of the ionophore A23187, under diverse metabolic conditions, and the results were used to interpret the adjustments required for good model fits. By the inclusion of low-affinity Mg2+ binding to ATP and bisphosphoglycerate, and also binding of Mg2+ to haemoglobin (four ions per tetramer) with an apparent dissociation constant of 45 mM we were able to realistically model, for the first time, all the experimentally observed changes in [Mg2+]i in human red cells under diverse metabolic conditions.