Arc front stratovolcanoes have global chemical systematics that constrain processes at convergent margins. Positive correlations exist for arc averages among ''fluid mobile,'' ''high field strength,'' and ''large ion lithophile'' elements. 143 Nd/ 144 Nd and 87 Sr/ 86 Sr from rear-arc lavas lacking subduction signature align with the oceanic ''mantle array,'' and correlate with arc front 143 Nd/ 144 Nd. Most chemical parameters (but not isotopes) also correlate well with crustal thickness and slightly less well with the slab thermal parameter, but not with the depth of the slab nor model slab surface temperatures. Successful models of arc volcanism should account for these global regularities. Two distinct models can quantitatively account for the observationsdifferent extents of melting of the mantle wedge caused by variations in wedge thermal structure, or varying contributions from the subducting slab owing to variations in the slab thermal structure. Both successful model scenarios require a significant flux of melted ocean crust to the mantle source of all volcanic arcs. The wedge melting model has constant contributions from ocean crust, sediment, and mantle wedge to lavas globally, while the slab model varies slab contributions with slab temperature. The wedge melting model fit improves by incorporating convergence rate and slab dip, which should affect the wedge thermal structure; the slab model is not supported by a similar analysis. The wedge model also more easily accommodates the isotope data. The two models predict different primary H 2 O contents, with large variations in H 2 O for the wedge model, and relatively constant H 2 O for the slab model. An evaluation of the effects of varying sediment compositions on arc lavas will benefit from considering the very different consequences of the two models.