A combined study of zircon U–Pb ages, whole-rock chemistry, Sr–Nd isotopes and in situ Lu–Hf isotopic ratios in zircons was carried out on Permian monzogranites and mafic microgranular enclaves (MME) and coeval massive gabbros in the northern Alxa block, North China. The data obtained were used to constrain magma sources and petrogenetic processes involved in the generation of this igneous suite. Zircon U–Pb dating yields ages of 271 ± 1 Ma, 270 ± 1 Ma and 276–270 Ma for the monzogranites, MME and massive gabbros, respectively. Two populations of MME, gabbroic (SiO2 <49 wt %) and dioritic (SiO2 >53 wt %) enclaves, are identified. They represent mafic to intermediate magmas quenched in a partially crystallized granitic (sensu lato) host; evidence to support this conclusion includes their fine-grained textures, sinuous margins and diffuse contacts with the host monzogranites. Back-veins and xenocrysts of quartz and plagioclase, as well as various disequilibrium textures and mineral assemblages, indicate mingling or mixing processes. The two magma systems, mafic and felsic, have broadly similar isotopic characteristics with whole-rock initial 87Sr/86Sr(i) ratios ranging from 0·7075 to 0·7077 in the monzogranite host and from 0·7067 to 0·7069 in the MME, with εNd(t) values ranging from –10·2 to –12·4 in monzogranites and from –8·2 to –9·9 in the MME. Zircon εHf(t) values of the monzogranites, gabbroic enclaves and massive gabbros show a wide range and significant overlap from –1·2 to –15·2, –4 to –13·3 and –5·4 to –19·5, respectively; the maximum frequency value of the Hf model age is almost coincident with the whole-rock Nd model age. Mafic, gabbroic rocks similar in composition to some enclaves form layered synplutonic intrusions several metres in thickness and more than 100 m in lateral extent. A mixing test based on mass balance for whole-rock major element compositions reveals that mixing was an efficient process between two coeval magmas. The fraction of felsic magma involved in the hybrid rocks ranged from 0·19–0·29 in the dioritic enclaves to 0·84 in granodiorite and good linear fits (r2 > 0·9) are obtained using the average composition of gabbroic enclaves and the most felsic monzogranite as end-members. Although the monzogranites and enclaves may be derived from distinct magma sources, they share similar isotopic signatures, pointing to interaction processes in the source region. The combination of information from geochronology, petrology, whole-rock geochemistry and isotopic compositions leads us to conclude that the processes of hybridization and magma mixing were effective at both the level of emplacement in the shallow crust and at depth in the magma source region. Hybridization in a source region within the lithospheric mantle, involving mantle and crustal source rocks, produced magma bimodality with strong geochemical affinities between end-members and clear calc-alkaline arc signatures, compatible with a subduction setting. A plausible subduction erosion plus relamination model, which differs from classical models based on mafic magma underplating, is proposed. Accordingly, the Yamatu monzogranites are argued to have been generated from granitic melt segregated and ponded in buoyant silicic diapirs, which formed by melting of subducted mélanges in the lithospheric mantle and eventually relaminated to the lower crust. They represent the melts that metasomatized the mantle region during a pre-Permian subduction event. The massive gabbros were formed by decompression melting of the previously metasomatized mantle. The gabbroic enclaves and gabbro layers of the Yamatu pluton are interpreted as magmas formed by decompression melting of this modified hydrated mantle. The dioritic enclaves represent hybrid liquids generated from reaction between granitic melts, which were derived from subducted mélanges, and the hydrated mantle or melts derived from it.