We performed a thermodynamic and experimental study to investigate the fate of pyroxenite-derived melts during their migration through the peridotitic mantle. We used a simplified model of interaction in which peridotite is impregnated by and then equilibrated with a finite amount of pyroxenite-derived liquid. We considered two pyroxenite compositions and three contexts of pyroxenitic melt impregnation: (1) in a subsolidus lithospheric mantle; (2) beneath a mid-ocean ridge (MOR) in a subsolidus asthenospheric mantle at high pressure; (3) beneath a MOR in a partially molten asthenospheric mantle. Calculations were performed with pMELTS at constant pressure and temperature with a melt–rock ratio varying in the range 0–1. Concurrently, a series of impregnation experiments was performed at 1 and 1·5 GPa to reproduce the final stages of the calculations where the melt–rock ratio is 1. Incoming melt and host-rocks react differently according to the melt composition and the physical state of the surrounding mantle. Whereas clinopyroxene (Cpx) is systematically a reaction product, the role of olivine (Ol) and orthopyroxene (Opx) depends on the incoming melt silica activity a^0_(SiO2): if it is lower than the silica activity Formula of a melt saturated in Ol and Opx at the same pressure P and temperature T, Opx is dissolved and Ol precipitates, and conversely if a_(SiO2) > a^0_(SiO2). Such contrasted reactions between pyroxenitic melts and peridotitic mantle may generate a large range of new lithological heterogeneities (wehrlite, websterite, clinopyroxenite) in the upper mantle. Also, our study shows that the ability of pyroxenite-derived melts to migrate through the mantle depends on the melting degree of the surrounding peridotite. The reaction of these melts with a subsolidus mantle results in strong melt consumption (40–100%) and substantial Cpx production (with some spinel or garnet, depending on P). This is expected to drastically decrease the system permeability and the capacity of pyroxenite-derived melts to infiltrate neighbouring rocks. In contrast, melt migration to the surface should be possible if the surrounding mantle is partially melted; although liquid reactivity varies with composition, melt consumption is restricted to less than 20%. Hence, magma–rock interactions can have a significant impact on the dynamics of melting and magma migration and should not be neglected when modelling the partial melting of heterogeneous mantle.