A detailed analysis of the petrographic relations and mineral chemical zoning combinedwith the use of calculated psuedosections is a powerful tool to unravel cryptic, yetimportant details of the P–T evolution of rocks, with a potential impact on theinterpretation of the geodynamic context.A mafic kyanite-bearing eclogite from the Haut-Allier displays garnet-rich layers thathave intriguing similarities and differences with respect to the surrounding eclogites.Chemical zoning of garnet inside and outside the layers is similar, but the layers areenriched in rutile and ilmenite, and lack kyanite, common in the eclogite. Threesuccessive metamorphic stages, M1, M2 and M3, are characterized by assemblagescomprising garnet1-omphacite- kyanite, garnet2-plagioclase, and amphiboleplagioclase,respectively, and define a clockwise P–T path. These events occurred atthe conditions of eclogite (M1; ∼ 20 kbar, 650 °C to ∼22.5 kbar, 850 °C), high-pressure granulite (M2; 19.5 kbar and 875 °C) and hightemperatureamphibolite facies (M3; < 9 kbar, 750– 850 °C), respectively. Phasediagrammodelling of garnet growth zoning and mineralogy of the inclusions reveal aprograde M1 stage, first dominated by burial and then by near-isobaric heating.Subsequent garnet1 resorption, prior to a renewed growth of garnet2 is interpreted interms of a decompression during M2. High- pressure partial melting is predicted forboth the M1 temperature peak and M2, due to focussed influx of H2O-rich fluids orexternal melts, probably associated with localized deformation. It is inferred thatsubsequent melt loss resulted in the formation of the garnet- rich layers. Despite theabsence of clear accumulations of crystallised melts in the eclogite, high-pressuremelting may he linked to the existence of HP trondhjemitic melts, described elsewherein the Massif Central. M3 testifies to further strong decompression associated withlimited cooling. The preservation of garnet growth zoning indicates the short-livedcharacter of the temperature increase, decompression and cooling cycle. We argue thatsuch P–T evolution is compatible with the juxtaposition of the asthenosphere againstthe subducted crust prior to exhumation driven by slab rollback.