Instrumental-neutron-activation analyses are reported for two uncontaminated rocks, a phlogopite-rich clot, and two contaminated rocks from the Coyote Peak diatreme, northwestern California. These data, combined with Nd, Sr, and Pb isotopic evidence, have been modeled to a multi-stage evolution for the uncontaminated rocks. Fertile mantle material (refractory elements 2.5× chondritic abundances; Rb/Sr = 0.029 by weight) was depleted about 900 m.y. ago by congruent melting and removal of ~4% basaltic liquid; this depleted residue provided the source rock from which the Coyote Peak magma was ultimately derived. About 66 m.y. ago, the depleted mantle residue was incongruently melted in the presence of H₂O and CO₂ at a total pressure > 26 kb to yield ~0.5% of a Si-poor, Ca-rich melt. This melt then metasomatized depleted garnet-free harzburgite in the upper mantle at about 26 kb to produce a rock similar to phlogopite-bearing wehrlite. About 29 m.y. ago, this rock was subjected to an increase in pressure to >26 kb and incongruently melted to give ~0.5% of a second-stage melt resembling olivine melilitite in composition. Enroute to the surface, about 28% olivine and 2% titanomagnetite were lost from the highly fluid melt.

Coarse-grained phlogopite-rich clots in the uncontaminated rocks apparently crystallized from a latestage liquid derived from the uncontaminated melt. Contaminated rocks appear to be the result of partial assimilation of, and dilution by, ~14% Franciscan graywacke country rock.

The diatreme was emplaced near a converging plate margin where young hot oceanic mantle and crust of the Juan de Fuca plate was probably subducting obliquely beneath a thin lip of the North American plate. The unusual chemistry of the rocks may be the result of this complex tectonic setting which could also have included local strike-slip and extensional environments within the two plates pierced by the diatreme.