The Eocene Tamazeght complex, High Atlas Mountains, Morocco is a multiphase alkaline to peralkaline intrusive complex. A large variety of rock types (including pyroxenites, glimmerites, gabbroic to monzonitic rocks, feldspathoidal syenites, carbonatites and various dyke rocks) documents a progression from ultramafic to felsic magmatism. This study focuses on the silicate plutonic members and the genetic relationships between the various lithologies. Based on detailed petrographic and mineral chemical data we show that the various units crystallized under markedly different oxygen fugacity and silica activity conditions and demonstrate how these parameters influence both the phase assemblage and the detailed chemical evolution of the fractionating phases. Nepheline, olivine–clinopyroxene and hornblende–plagioclase thermometry indicate equilibration temperatures ≥800°C for all major rock types. Highly oxidized conditions (close to the hematite–magnetite buffer) are characteristic of the garnet-rich pyroxenites, ultrapotassic glimmerites and associated olivine-shonkinites. The parental magmas to these rocks evolved from low initial aSiO2 values of 0·1 to values of 0·5–0·8 during nepheline and alkali feldspar saturation. In contrast, the monzonitic rocks evolved from initially high aSiO2 values (up to 0·75) down to about 0·1 at intermediate values of oxygen fugacity (ΔFMQ=+2–5 to −1, where FMQ is the fayalite–magnetite–quartz buffer). For nepheline syenites and malignites, more reduced conditions (ΔFMQ=−2) and intermediate aSiO2 values (between 0·25 and 0·5) dominate. We conclude that fractional crystallization is not a likely mechanism to explain the large variety of lithologies present in the Tamazeght complex. It is more probable that successive melting of a compositionally heterogeneous mantle source region gave rise to several melt batches with distinct chemical and physico-chemical characteristics. Low-degree melts from a K-phase-bearing mantle domain resulted in the formation of ultrapotassic glimmerites, whereas garnet-rich pyroxenites and olivine-shonkinites may have originated from hybrid melts and partly from a pyroxene-dominated source. Less alkaline lithologies such as monzonites potentially reflect larger degrees of melting and the increased importance of a basaltic component, whereas nepheline syenites and malignites may be explained by lower degrees of melting and a more alkaline character for the parental melt of these rocks.