New whole-rock and Sr–Nd isotopic analyses of Quaternary lavas from Kurdistan Province, western Iran, shed light on the nature of orogenic plateau magmatism during continental collision and the possible role of accessory minerals during mantle partial melting in this tectonic setting. The sampled lavas are from the Turkish–Iranian plateau within the Arabia–Eurasia collision zone. Compositions are typically basanite, hawaiite and alkali basalt, with minor rhyolite. Most of the basic samples from the Qorveh–Bijar region have elevated abundances of large ion lithophile elements (LILE) and light rare earth elements (LREE) (e.g. 76 ppm < La < 165 ppm), with steep REE patterns: 45 < La/Yb < 101. Two flows from Takab in the north of the study area have higher K2O/Na2O and Rb than the Qorveh–Bijar samples, but lower LREE and Ba. Sr–Nd isotope values for all the basic samples plot close to Bulk Silicate Earth, with 143Nd/144Nd between 0·51263 and 0·51276, and 87Sr/86Sr between 0·70467 and 0·70600; this is similar to Quaternary alkali basalts from the Iran–Turkey borderlands to the NW, but distinct from a more depleted source melting elsewhere in the same collision zone; for example, at Mount Ararat and the Kars plateau (Turkey). Crustal contamination does not appear to be an important factor affecting magma composition. The range of chemical signatures suggests variable melting of at least two distinct sources. One inferred source produced melts with La/Nb ranging from ∼3·5 to ∼1·2, which is unusual for volcanic rocks that are otherwise coeval. We interpret this variation as the result of depletion of a K-richterite- and rutile-bearing source during melting in the garnet stability field. The importance of this result is that the size of the negative Nb anomaly and fractionation of LILE may depend on K-richterite and rutile in the source mineralogy, rather than simply the bulk abundances of incompatible trace elements in the source region. We infer the presence of phlogopite in a second mantle source, the melting of which produced the more potassic lavas from Takab. Lithosphere delamination or slab break-off mechanisms for triggering melting are problematic in the study area, as the lithosphere is reportedly ∼150–200 km thick. Late Cenozoic extension has not been recognized, and so extension is unlikely as a cause of melting. The NW–SE alignment of volcanic centres, parallel to the regional structural grain, does imply a structural control, at least on the final eruption site. Eurasian mantle lithosphere was probably metasomatized by fluids derived from Mesozoic–early Cenozoic subduction and the early stages of collision. More recent convergence has had the potential to cause further fluxing and melting of the re-fertilized lithosphere, with rapid magma ascent assisted by transcurrent motion along the orogen. It is also possible that the negative dT/dP section of the amphibole peridotite solidus was crossed as a result of lithospheric thickening in the collision zone (i.e. compression melting rather than decompression melting). Exhaustion of accessory phases may also help explain larger-scale transitions from arc to within-plate chemistry in orogens, such as the Paleogene Arabia–Eurasia system, and the Mesozoic–Cenozoic magmatism of eastern China.