Primitive late Cenozoic basaltic lavas from the Cascades volcanic arc near latitude 46°N comprise two distinct compositional groups. Group I includes samples with low Ba/Nb (<20) and other compositional similarities to oceanic island and MORB lavas from within-plate settings. In contrast, Group II exhibits enrichment of Ba and large-ion lithophile elements (LILE) and depletion of Nb and high-field strength elements (HFSE) as seen commonly in calcalkalic lavas from other volcanic arcs. Lavas of both groups are widely distributed across the transect, and Group I lavas are found as much as 30–40 km trenchward of stratovolcanoes that define the High Cascades ‘volcanic front (VF)’. The most primitive lavas are sparsely porphyritic, have elevated Ni, Cr, and Mg#, high calculated magmatic temperatures (1200–1300 °C), and lack evidence of shallow (crustal level) storage and crystallization. Compositions of parental liquids were calculated for each primitive sample on the premise of Fe–Mg equilibrium with mantle peridotite. Assuming that such magmas ascended rapidly from accumulation zones in the mantle, we estimate P and T of segregation. We infer that (a) Group I magmas ascended from systematically greater depths (∼50–70 km) than Group II (∼30–50 km), implying the possible existence of compositional stratification in the mantle wedge; (b) Group I basalts show the least evidence for slab-derived contributions in their sources despite their apparently greater segregation depths (approaching the locus of the Cascadia slab beneath the frontal arc region); (c) Group II lavas with the strongest slab compositional signature have temperatures far exceeding the wet peridotite solidus at high pressure; and (d) the inferred thermal structure of the mantle wedge is very warm, implying a significant component of mantle upwelling and convection. Group I lavas are interpreted as decompression melts from this mantle, and their compositions suggest that their source was little modified by slab-derived contributions. We speculate that melting to produce Group II magmas occurs in the shallow mantle, possibly in response to heating by hot ascending Group I magmas. If true, it seems unlikely that the slab-like signal in Group II lavas can be attributed to modern slab inputs; rather, we postulate that this signature may reflect melting of lithospheric mantle domains containing a ‘stored’ slab-derived component inherited from earlier stages of Cascadia subduction. This scenario differs from the standard paradigm for subduction zones (SZs), and stresses the importance of convecting asthenospheric mantle in driving arc magmatism, particularly in warm subduction zones where slab fluid contributions likely are minimal. In contrast, because tectonic conditions in more typical volcanic arcs favor subduction of cooler, less dehydrated oceanic lithosphere, slab-derived fluids may promote extensive flux-melting in the wedge. Such melts may dominate the magmatic output and mask wedge contributions. The Cascade arc thus provides rarely afforded insights into arc magma genesis.