Calculations on a new empirical solution model, constructed as a Kohler oxide solution with special oxide coefficients and nearly symmetrical binary interaction parameters, reproduce the simple geometry found for silicate liquid immiscibility in synthetic systems. Both ordinary regular and subregular solutions are unsuitable for these calculations because, as components are added, these solutions result in complicated liquid immiscibility with many extraneous solvii. The geometry of immiscibility is determined by spinodal calculation followed by a graphical construction. Spinodal calculation produces a vector R which must be roughly parallel to tie lines and is used to compare the real with modelled partitioning. The results agree very well with the available experimental work for SiO2- M(1) O- M(2) O systems M = Fe, Mg, Ca, Mn, Zn, Ba, Pb) and reasonably well for the quartz-fayalite-leucite section with minor Fe 2O 3, TiO 2 and P 2O 5. The new model predicts that the flattened liquidus of diopside, in the diopside-leucite-quartz and diopside-nepheline-quartz systems, is due to the metastable extension of the diopside-quartz melt solvus, only 100-150°C below the liquidus of diopside. Preliminary attempts to extend the coefficient model to natural examples of magma immiscibility are not very successful.