In the world-famous sediments of the Chinese Loess Plateau, fossil soils alternate with windblown dust layers to record monsoonal variations over the last ∼3 My. The less-weathered, weakly magnetic dust layers reflect drier, colder glaciations. The fossil soils (paleosols) contain variable concentrations of nanoscale, strongly magnetic iron oxides, formed in situ during the wetter, warmer interglaciations. Mineralogical identification of the magnetic soil oxides is essential for deciphering these key paleoclimatic records. Formation of magnetite, a mixed Fe2+/Fe3+ ferrimagnet, has been linked to soil redox oscillations, and thence to paleorainfall. An opposite hypothesis states that magnetite can only form if the soil is water saturated for significant periods in order for Fe3+ to be reduced to Fe2+, and suggests instead the temperature-dependent formation of maghemite, an Fe3+-oxide, much of which ages subsequently into hematite, typically aluminum substituted. This latter, oxidizing pathway would have been temperature, but not rainfall dependent. Here, through structural fingerprinting and scanning transmission electron microscopy and electron energy loss spectroscopy analysis, we prove that magnetite is the dominant soil-formed ferrite. Maghemite is present in lower concentrations, and shows no evidence of aluminum substitution, negating its proposed precursor role for the aluminum-substituted hematite prevalent in the paleosols. Magnetite dominance demonstrates that magnetite formation occurs in well-drained, generally oxidizing soils, and that soil wetting/drying oscillations drive the degree of soil magnetic enhancement. The magnetic variations of the Chinese Loess Plateau paleosols thus record changes in monsoonal rainfall, over timescales of millions of years.