Physical and chemical constraints require that Columbia River basalt flows were fed from a large (>700 km3) magma reservoir near the crust-mantle boundary and prohibit significant crystal fractionation or crustal assimilation between reservoir and surface. Demonstrable fractionation of plagioclase + olivine ± pyroxene in the reservoir largely obviates the need for an unusual iron-rich pyroxenite source. Variation of Sr and Nd isotope ratios with age in the main series of Columbia River basalt and variation in such incompatible trace-element ratios as Ba/P suggest lower crustal assimilation at the top of the reservoir combined with crystal fractionation. However, decoupling between isotope ratios and incompatible element ratios implies two separate processes in addition to crystal fractionation, of which lower crustal assimilation may be one. The other process may involve the participation of a vertically heterogeneous mantle resulting from metasomatism in the continental lithosphere.