<p>Zircon is a powerful pathfinder mineral for identifying igneous source rocks, including kimberlite and carbonatite. However, discrimination of zircons from these two lithologies is challenging due to their surprisingly limited published trace element data. Zircon megacrysts from kimberlite can have a wide range of U–Pb ages, from Archean to Eocene in some locations, raising questions about their origin. Here, we determined the trace-element compositions of 170 new zircon megacrysts from kimberlites, four from carbonate-rich olivine lamproites, five from ultramafic lamprophyres, one from a lamprophyre, and two from a mica-amphibole-rutile-ilmenite-diopside (MARID) xenolith. We also determined the trace-element compositions of 220 new zircons from global carbonatites and related rocks. The kimberlitic zircons in the present study are all megacrysts with a relatively narrow range of trace-element compositions whereas the new carbonatite zircons are compositionally diverse and likely reflect formation under varied geological conditions from a variety of heterogeneous sources, as well as complex equilibrium mineral assemblages. We apply random forest (RF) and discriminant projection analysis (DPA) to distinguish zircons from kimberlite and carbonatite from those in many crustal lithologies. DPA-based graphical methods employ these elements to allow rapid evaluation of zircon provenance using elemental data with an intuitive and interpretable interface. We further apply our compiled database to attempt to search for a compositional fingerprint, using machine learning, that might be capable of distinguishing megacryst zircons from kimberlites containing superdeep diamonds from those that do not. We provide a Microsoft Excel workbook for the rapid classification of zircons using DPA, and an R-based software package for the classification of zircons using RF.</p>