Knowledge of melting properties is critical to predict the nature and the fate of melts produced in the deep mantle. Early in the Earth’s history, melting properties controlled the magma ocean crystallization, which potentially induced chemical segregation in distinct reservoirs. Today, partial melting most probably occurs in the lowermost mantle as well as at mid upper-mantle depths, which control important aspects of mantle dynamics, including some types of volcanism. Unfortunately, despite major experimental and theoretical efforts, major controversies remain about several aspects of mantle melting. For example, the liquidus of the mantle was reported (for peridotitic or chondritic-type composition) with a temperature difference of ∼1000 K at high mantle depths. Also, the Fe partitioning coefficient (DFeBg/melt) between bridgmanite (Bg, the major lower mantle mineral) and a melt was reported between ∼0.1 and ∼0.5, for a mantle depth of ∼2000 km. Until now, these uncertainties had prevented the construction of a coherent picture of the melting behavior of the deep mantle.In this article, we perform a critical review of previous works and develop a coherent, semi-quantitative, model. We first address the melting curve of Bg with the help of original experimental measurements, which yields a constraint on the volume change upon melting (ΔVm). Secondly, we apply a basic thermodynamical approach to discuss the melting behavior of mineralogical assemblages made of fractions of Bg, CaSiO3-perovskite and (Mg,Fe)O-ferropericlase. Our analysis yields quantitative constraints on the SiO2-content in the pseudo-eutectic melt and the degree of partial melting (F) as a function of pressure, temperature and mantle composition; For examples, we find that F could be more than 40% at the solidus temperature, except if the presence of volatile elements induces incipient melting. We then discuss the melt buoyancy in a partial molten lower mantle as a function of pressure, F and DFeBg/melt. In the lower mantle, density inversions (i.e. sinking melts) appear to be restricted to low F values and highest mantle pressures.The coherent melting model has direct geophysical implications: (i) in the early Earth, the magma ocean crystallization could not occur for a core temperature higher than ∼5400 K at the core-mantle boundary (CMB). This temperature corresponds to the melting of pure Bg at 135 GPa. For a mantle composition more realistic than pure Bg, the right CMB temperature for magma ocean crystallization could have been as low as ∼4400 K. (ii) There are converging arguments for the formation of a relatively homogeneous mantle after magma ocean crystallization. In particular, we predict the bulk crystallization of a relatively large mantle fraction, when the temperature becomes lower than the pseudo-eutectic temperature. Some chemical segregation could still be possible as a result of some Bg segregation in the lowermost mantle during the first stage of the magma ocean crystallization, and due to a much later descent of very low F, Fe-enriched, melts toward the CMB. (iii) The descent of such melts could still take place today. There formation should to be related to incipient mantle melting due to the presence of volatile elements. Even though, these melts can only be denser than the mantle (at high mantle depths) if the controversial value of DFeBg/melt is indeed as low as suggested by some experimental studies. This type of melts could contribute to produce ultra-low seismic velocity anomalies in the lowermost mantle.