Abstract Carbonatites are rare igneous rocks that host the largest resources of REE and Nb, yet, their genesis and evolution are far from clear. The leading models of carbonatite formation are the direct melting of carbonate-bearing peridotites, silicate-carbonatite liquid immiscibility, and fractionation of carbonated silicate melts. The validity of these models has never been robustly addressed through combining the available experimental results with the natural rock record. We thus re-evaluate the presently 633 alleged carbonatite occurrences including carbonatite type, bulk composition, mineralogy, and field exposure, followed by a review of experimental data pertinent to carbonatite genesis and evolution. Based on the available data, 454 carbonatite occurrences are magmatic, of which 87 without and 338 with spatially associated alkaline magma, 9 with kimberlites, and 20 with ultramafic cumulates only. Eighty-four percent of the magmatic occurrences contain calcite carbonatite (of which 1/3 also contain dolomite carbonatite), only 9% have dolomite but not calcite carbonatite, the incidence of dolomite carbonatite being similar for occurrences with or without associated silicate magmas. Available experimental data show that crystallization of calcite, dolomite, ankerite, and siderite at crustal conditions requires moderately alkaline and/or hydrous carbonate melts with ≥20 to 25 wt % (Na,K)2CO3 + H2O. It follows that carbonatite rocks, poor in these elements, are at best magmatic cumulates (if not carbo- or hydrothermal) that lost these ephemeral components. Carbonatitic melts could form in the lithospheric mantle, but these are always dolomitic and cannot deviate from close-to-minimum compositions when rising, their strong adiabatic cooling keeping them on the solidus until they decompose to olivine, clinopyroxene (cpx), and CO2 when reaching <2.1 GPa, i.e. the carbonated peridotite solidus ledge, which renders their extraction from the mantle highly unlikely. Furthermore, dolomitic carbonate melts crystallize periclase + calcite at crustal conditions. Only when containing ≥15 wt % (Na,K)2O + H2O they may crystallize dolomite and form dolomitic carbonatites. This value is far above the 2 to 5 wt % (Na,K)2O of mantle-derived carbonatitic melts. Liquid immiscibility from CO2-bearing close-to-natural melilititic, nephelinitic, and phonolitic melts requires 10 to 15 wt % Na2O + K2O in the silicate melt, increasing with SiO2. Extensive differentiation of primitive alkaline melilititic or basanitic parents is hence required to achieve immiscibility. The experimental data show that evolved nephelinites and phonolites unmix calcic carbonatitic melts, while melilitites and undifferentiated nephelinites with >4 wt % MgO may also unmix dolomitic carbonatitic melts. The latter may hence arise from liquid immiscibility or develop through fractionation from calcic carbonatitic melts. Finally, carbonatites may also derive through fractionation of CO2-rich ultramafic melts, but a continuous increase in dissolved CO2 from a carbonated silicate melt to a carbonatitic melt requires ≥3 GPa. We conclude that the combination of the natural rock record with melting and crystallization phase relations excludes a direct mantle origin for almost all carbonatites found in the crust. Instead, their vast majority forms through immiscibility from an alkali-rich differentiated silicate melt that stems from a mantle-derived alkaline parent, consistent with the common spatial association with alkaline complexes and similar isotopic compositions of carbonatite and alkaline silicate rocks. Direct fractionation from silicate melts may occur for kimberlitic or ultramafic lamprophyric melts, but only at ≥3 GPa, i.e. within the lithospheric mantle. To make progress in this field, we suggest a more rigorous distinction of magmatic and carbo- or hydrothermal carbonatite rocks in each occurrence, and to focus on mineral compositions in the carbonatite and associated silicate magmas, as bulk rocks are at best cumulative in nature. Additional experimental work to understand the role of alkalis and H2O in the formation and evolution of carbonatites, in particular crystallization and fluid saturation at crustal conditions, will be essential to provide a more complete understanding of carbonatite petrogenesis.