We present a detailed study of two ca. 200 m-thick apatite-bearing ferrogabbro horizons of the Sept Iles layered intrusion (Canada). These rocks are the most evolved cumulates of the megacyclic units (MCU) I and II, and mark the transition between basaltic and silicic magmatism. They are made up of plagioclase (An 55-34 ), olivine (Fo 66-21 ), clinopyroxene (Mg#75-55), ilmenite, magnetite, apatite, ± pigeonite and are a significant source of Fe-Ti-P ore. Ferrogabbros have relatively uniform bulk-rock compositions in MCU I but are bimodal in MCU II. The liquid lines of descent for major elements in equilibrium with cumulates of MCU I and II have been calculated using a forward model formalism. Both trends evolve towards SiO 2 -enrichment and FeO t -depletion after saturation in Fe-Ti oxides. However, because of magma mixing in MCU II, they do not follow the same path. Evolved liquids from MCU II are shown to enter the experimentally-determined two liquid stability field, while MCU I liquids do not. Immiscibility in MCU II and its absence in MCU I are supported by the presence of contrasted reactive symplectites in cumulate rocks. Apatite-bearing ferrogabbros in MCU II have crystallized from distinct immiscible Fe-rich and Si-rich silicate melts which have physically segregated in the slow-cooling magma chamber. Two different types of cumulate rocks are thus produced: leucocratic and melanocratic gabbros. This is consistent with the presence of Si-rich and Fe-rich melt inclusions in apatite. In contrast, homogeneous ferrogabbros from MCU I were produced by simple fractional crystallization of a homogeneous liquid. Our data suggest that immiscibility could also explain the large geochemical variability of ferrogabbros in the Upper Zone of the Bushveld Complex (South Africa).