We report major and trace element abundances for 147 samples and Sr, Nd, Hf, and Pb isotope compositions for a 36 sample subset of basaltic lava flows, sills, and dykes from the Karoo continental flood basalt (CFB) province in Botswana, Zimbabwe, and northern South Africa. Both low- and high-Ti (TiO2 < 2 wt % and > 2 wt %) rocks are included. MELTS modeling shows that these magmas evolved at low pressure (1 kbar) through fractional crystallization of gabbroic assemblages. Whereas both groups display enrichment in light rare earth elements (LREE) relative to heavy REE (HREE) and high field strength elements, and systematic negative Nb anomalies, they differ in terms of contrasting middle REE (MREE) to HREE fractionation, which is greater for the high-Ti basalts. This reflects different depths of melting of slightly enriched mantle sources: calculations suggest that the low-Ti basalts were generated by melting of a shallow spinel-bearing (2 % spinel) lherzolite, whereas the high-Ti magmas originated from a deeper-seated garnet-bearing (2–7% garnet) lherzolite. In most isotope plots, the high-Ti lavas together with the picrites define a common trend from Bulk Silicate Earth (BSE) to compositions with strongly negative ɛNdi and ɛHfi akin to those of some nephelinites and lamproites. The low-Ti rocks are shifted from BSE-like to more radiogenic Sr isotope ratios, indicative of upper crustal contamination. Trace element and isotope characteristics of the Karoo magmas require a combination of enrichment processes (subduction induced?) and long-term isolation of the mantle sources. We propose two distinct scenarios to explain the origin of the Karoo province. The first calls for polybaric melting of spatially heterogeneous, partially veined, sub-continental lithospheric mantle (SCLM). Calculations show that mixing between SCLM (∼BSE) and a strongly Nd–Hf unradiogenic nephelinite-like component (sediment input?) could account for the compositional variations of most of the high-Ti group lavas, whereas the mantle composition responsible for the low-Ti magmas is more likely to be similar to a vein-free, metasomatically enriched SCLM component. The second scenario involves mixing between two end-members represented by the SCLM and its deep-seated alkalic veins and a sub-lithospheric (asthenospheric- or ocean island basalt-like?) mantle plume. In this case, the data are compatible with an increasing mantle plume contribution as the plume rises and expands through the lithosphere. Regardless of which of the two scenarios is invoked, the spatial distribution of the low- and high-Ti magmas matches the relative positioning of the cratons and the Limpopo belt in such a way that strong control of the lithosphere on magma composition and distribution is a mandatory requirement of any petrogenetic model applied to the Karoo CFB.