Abstract The Duluth Complex is a large mafic intrusive system located in northeastern Minnesota emplaced as part of the 1.1-Ga Midcontinent Rift. Several Fe–Ti oxide-bearing ultramafic intrusions are hosted along the Western Margin of the Duluth Complex, and are discordant bodies present in a variety of geometries, hosted in multiple rock types, and dominated by peridotite, pyroxenite, and semi-massive to massive Fe–Ti oxide rock types. Their origin has been debated, and here we present geochemical evidence and modeling that supports a purely magmatic origin for the Titac and Longnose Fe–Ti oxide-bearing ultramafic intrusions. Ilmenite and titanomagnetite textures indicate a protracted cooling process, and δ34S values of sulfides reveal little assimilation of the footwall Virginia Formation, a fine-grained pelitic unit that contains sulfide-rich bands. We model the crystallization of a hypothetical parental magma composition to the host intrusion of Longnose using Rhyolite-MELTS and demonstrate that the accumulation of Fe–Ti oxides in the discordant intrusions cannot be explained by density-driven segregation of crystallized Fe–Ti oxides. Instead, we show that the development of silicate liquid immiscibility, occurring by the unmixing of the silicate melt into conjugate Si- and Fe-rich melts, can result in the effective segregation and transportation of the Fe-rich melt. The Fe-rich melt is ~2 orders of magnitude less viscous than the Si-rich melt, allowing the Fe-rich melt to be more effectively segregated and transported in the mush regime (crystallinities >50%). This suggests that viscosity, in addition to density, plays a significant role in forming the discordant Fe–Ti oxide-bearing ultramafic intrusions. We propose a genetic model that could also be responsible for the Fe–Ti oxide-rich layers or bands that are hosted within the igneous stratigraphy of mafic intrusions of the Duluth Complex.