Megacrysts are large (cm to >20 cm in size) mantle-derived crystals, which are commonly entrained by kimberlite magmas, comprising of olivine, orthopyroxene, clinopyroxene, phlogopite, garnet, ilmenite and zircon as common phases. Numerous studies have shown megacrysts to contain polymineralic inclusions, which have been interpreted to represent entrapped kimberlite melt. To constrain the origin of these inclusions in megacrysts and their relationship to kimberlite magmatism, we present a detailed petrographic and geochemical study of clinopyroxene and olivine megacrysts and their hosted inclusions from the Diavik, Jericho, Leslie (Slave Craton, Canada) and Udachnaya-East (Siberian Craton, Russia) kimberlites. The studied megacrysts are between 1 and 3 cm in size and representative of both the Cr-rich and Cr-poor suites. Megacrysts contain two types of inclusions: i. Large (<0.5–5 mm in size) round-to-irregular shaped polymineralic inclusions, which are composed of minerals similar to the host kimberlite groundmass, and consist of olivine, calcite, spinel, perovskite, phlogopite and apatite (± serpentine, alkali-carbonates, alkali-chlorides, barite). ii. Swarms/trails of ‘micro melt inclusions’ (MMI; <1–5 μm in size), which surround polymineralic inclusions, veins and fractures, thereby forming a ‘spongy’ texture. MMIs generally contain multiphase assemblages similar to polymineralic inclusions as well as various additional phases, such as alkali-carbonates or alkali-chlorides, which are typically absent in polymineralic inclusions and the surrounding kimberlite groundmass. Textural and geochemical evidence suggests that polymineralic inclusions in megacrysts crystallised from kimberlite melt, which infiltrated along fracture/vein networks. The polymineralic inclusion assemblages resulted from disequilibria reactions between the host megacryst and infiltrating kimberlite melt, which was likely enhanced by rapidly changing conditions during magmatic ascent. The connectivity of polymineralic inclusions to the kimberlite groundmass via network veins/fractures suggests that they are susceptible to infiltrating post-emplacement fluids. Therefore, the vast majority of polymineralic inclusions are unlikely to represent ‘pristine’ entrapped kimberlite melt. In contrast, MMIs are isolated within megacrysts (i.e. not connected to fractures/veins and therefore shielded from post-magmatic fluids) and probably represent entrapped remnants of the variably differentiated kimberlite melt, which was more enriched in alkalis-Cl-S-CO2 than serpentinised polymineralic inclusions and the host rocks exposed at Earth's surface as kimberlites.