The Stillwater igneous complex of Montana is an intrusion of lopolithic form. The floor and about 16,000 feet of layered rocks are exposed on the East Boulder plateau. It is estimated that another 40 per cent of the complex and its roof are hidden beneath a cover of Paleozoic rocks. North of the complex some Precambrian roof rocks are exposed by faulting and erosion. The order in which the essential minerals separated from the Stillwater magma is as follows: olivine, orthopyroxene, plagioclase, and clinopyroxene. Near the uppermost outcrops hypersthene inverted from pigeonite appears. The only other mineral present in rock-forming amount is the accessory, chromite, which is concentrated in a few layers in the lower part of the complex. The complex represents a natural example of fractional crystallization on a vast scale. Crystals separate from the magma and accumulate on the floor only to be covered by more precipitate which thereby prevents reaction with the remaining liquid. The chemistry of the essential minerals—plagioclase, orthopyroxene, and clino-pyroxene—is thoroughly investigated. New optical-property curves are presented, and the accuracy and suitability of various determinative procedures are discussed. Smith’s curves for low-temperature plagioclase represent a marked improvement over existing curves, and his determinative procedure is an advance both in precision and ease of measurement. The exsolution phenomena in the pyroxenes are used to interpret the subsolid relationships and to draw petrologic conclusions particularly as regards temperature during crystallization. Curves are drawn to show the order in which the essential minerals separated from magma and their composition changes as crystallization proceeded. The general trend of the curves from more magnesian to more iron-rich for the ferromagnesian minerals and from more calcic to more sodic for the plagioclases is evident. Of particular interest are the minor oscillations of these curves. These can be correlated with the effect of pore-space liquid on the accumulated crystals on the floor. The pore space in the crystal mush is at least 25 per cent by volume. The effect produced by this pore-space liquid is very much less than what would be expected; consequently it is postulated that diffusion from the pore-space liquid to the overlying main mass of magma and vice versa has tended to diminish the effect. When accumulation of crystals on the floor is slow diffusion may be so effective that the phases formed will not be appreciably different in composition from the accumulated crystals; when accumulation is rapid the pore-space liquid can have a large effect in modifying the composition of the crystals formed. Phases present in small amount among the accumulated crystals can be modified to a greater extent by pore-space liquid, and phases present in large amount are modified to a lesser extent. At any rate minor fluctuations in the mineral-variation curves can be correlated with the local rate of accumulation of crystals on the floor. The process also accounts for the formation of certain monomineralic rocks. Without diffusion the trapped pore-space liquid would have contributed small amounts of other phases to the rocks. With respect to layering in the Stillwater complex both the observations and conclusions closely parallel those of Wager and Deer for the Skaergaard intrusion. A theory to account for rhythmic layering is advanced. If crystals with different settling velocities are present and if the magma is in motion, rhythmic layering will result provided the vertical component of the motion in the magma is not constant. Velocities such as 100m/yr are of the right order to produce the desired result if the viscosity is 3000 poises. Well-developed rhythmic layering, however, is almost always associated with very large differences in settling velocities between phases. Thus it is conspicuously developed in the Stillwater complex, Norite Zone, where the ortho-pyroxene crystals are comparatively large and plagioclase comparatively small. The settling velocity difference is 25 to 1. This suggests that the process operates in a layer of magma very close to the floor, perhaps in the lower 100 meters. The composition of the magma is estimated from the composition of the border facies. It is very similar to the Bushveld complex. It is a basaltic magma, just saturated, and is comparatively low in TiO2, Fe2O3, and K2O. Otherwise it is much like the Karroo dolerites or Palisades diabase. An estimate of the composition of the hidden uppermost rocks is made by subtracting the composition of the exposed complex from the original magma in suitable proportions. If 40 per cent is hidden the resulting composition by subtraction is such that it corresponds to rocks of the uppermost portion of the Bushveld complex or Great Dyke—ferrogabbro. An estimate of the rate of heat loss during crystallization suggests that about 10 cm of crystals are deposited on the floor a year. If the roof is much thicker than 1 km this value may have to be reduced. Data on viscosity of basalts, densities of phases at high temperatures, and settling rates of crystals are analyzed. It is shown that two-phase convection in the magma is probable. In appendices certain general propositions are discussed. The calc-alkaline magma series so differs in its trends from those produced by fractional crystallization that some other origin seems required. A partial melting sequence is suggested. This leads to a consideration of the generation of magmas and the vertical composition gradient in continental crust. Differentiation of dolerite sills parallels the trend observed in the Stillwater complex but with a lower degree of fractionation. Absence of igneous lamination suggests that the process is unrelated to crystal settling. A process of inward crystallization from cooling surfaces analogous to the crystallization of the pore-space liquid in the Stillwater crystal mush is suggested; a diffusion mechanism removes the film of residual liquid.