One way to frame the debate about the relationships between volcanic and plutonic rocks is this: are plutons samples of magma that passed through the crust, or residues left behind by extraction of erupted liquids? In the former case plutons are compositionally equivalent to cogenetic volcanic rocks, barring biases introduced by passing through the crustal filter; in the latter they are cumulates, having lost liquid to eruption. These hypotheses make specific predictions about trace-element variations, which we test using global geochemical databases for circum-Pacific convergent margins and western North America. Volcanic rocks are far more abundant in these datasets than plutonic rocks and are biased to more mafic compositions. After subsampling the volcanic dataset to match the plutonic dataset, we find little evidence for significant loss of liquid from plutons. Rather, plutonic and volcanic trace-element patterns are generally indistinguishable. Where distinctions do occur, they are backwards; for example, a higher proportion of plutonic rocks has low Eu, Zr, and Ba, features of fractionated liquids, than volcanic rocks. These observations support the hypothesis that liquids fractionated from crystal-rich magmas are of small volume and are relatively immobile (e.g., aplites). These conclusions, derived from bulk-rock geochemistry, are supported by U-Pb zircon geochronology and field and textural observation. These data support the view that plutonic rocks are texturally modified samples of the same magmas that erupt. Partial melting provides an alternative to crystal fractionation for the origin of high-silica volcanic rocks.