The exhumation mechanism of high-pressure (HP) and ultrahigh-pressure (UHP) eclogites formed by the subduction of oceanic crust (hereafter referred to as oceanic eclogites) is one of the primary uncertainties associated with the subduction factory. The phase relations and densities of eclogites with MORB compositions are modelled using thermodynamic calculations over a P–T range of 1–4 GPa and 400–800 °C, respectively, in the NCKFMASHTO (Na2O–CaO–K2O–FeO–MgO–Al2O3–SiO2–H2O–TiO2–Fe2O3) system. Our modelling suggests that the mineral assemblages, mineral proportions and density of oceanic crust subducted along a cold P–T path are quite different from those of crust subducted along a warm P–T path, and that the density of oceanic eclogites is largely controlled by the stability of low-density hydrous minerals, such as lawsonite, chlorite, glaucophane and talc. Along a cold subduction P–T path with a geotherm of ~6 °C km−1, lawsonite is always present at 1.1 to >4.0 GPa, and chlorite, glaucophane and talc can be stable at pressures of up to 2.3, 2.6 and 3.6 GPa respectively. Along such a P–T path, the density of subducted oceanic crust is always lower than that of the surrounding mantle at depths shallower than 110–120 km (< 3.3–3.6 GPa). However, along a warm subduction P–T path with a geotherm of ~10 °C km−1, the P–T path is outside the stability field of lawsonite, and the hydrous minerals of chlorite, epidote and amphibole break down completely into dry dense minerals at relatively lower pressures of 1.5, 1.85 and 1.9 GPa respectively. Along such a warm subduction P–T path, the subducted oceanic crust becomes denser than the surrounding mantle at depths >60 km (>1.8 GPa). Oceanic eclogites with high H2O content, oxygen fugacity, bulk-rock XMg [ = MgO/(MgO + FeO)], XAl [ = Al2O3/(Al2O3 + MgO + FeO)] and low XCa [ = CaO/(CaO + MgO + FeO + Na2O)] are likely suitable for exhumation, which is consistent with the bulk-rock compositions of the natural oceanic eclogites on the Earth's surface. On the basis of natural observations and our calculations, it is suggested that beyond depths around 110–120 km oceanic eclogites are not light enough and/or there are no blueschists to compensate the negative buoyancy of the oceanic crust, therefore explaining the lack of oceanic eclogites returned from ultradeep mantle (>120 km) to the Earth's surface. The exhumed light–cold–hydrous oceanic eclogites may have decoupled from the top part of the sinking slab at shallow depths in the forearc region and are exhumed inside the serpentinized subduction channel, whereas the dense–hot–dry eclogites may be retained in the sinking slab and recycled into deeper mantle.