Compositional diversity among pristine nonmare rocks is probably too great for both the Mg-rich and ferroan anorthosite groups to be generated by a single magma, based on ratios of ''plagiophile'' elements (which have plagioclase/liquid distribution coefficients close to unity, but much smaller crystal/liquid distribution coefficients for olivine and low-Ca pyroxene, e.g., Al, Sr, Eu, Ga, and to lesser degrees K and Ba). Because plagiophile elements are virtually confined to a single mineral, bulk-rock ratios of plagiophile elements are nearly immune to sampling errors. On a plot of Eu/Al versus mg, ferroan anorthosites are separated from all other pristine nonmare rocks by a considerable gap. Stochastic processes (e.g., differential entrapment of interstitial liquid, or differential crystallization of minor phases) might conceivably explain the general diversity of plagiophile ratios, but a nonrandom process must be invoked to account for a gap in the spectrum of ratios. A single magma probably cannot account for even the Mg-rich pristine rocks subset, based on diversity of plagiophile ratios among samples with similar mg ratios. Plagiophile ratios also constrain the bulk composition of the moon. Previous work suggests that ferroan anorthosites formed by relatively uncomplicated processes, i.e., flotation of plagioclase cumulates over a primordial ''magmasphere.'' This model implies that ferroan anorthosite plagiophile ratios are straightforward related to those of the bulk fraction of the moon. For refractory elements, plagiophile ratios among ferroan anorthosites exactly match those expected under such a model, assuming the bulk moon ratios are approximately chondritic. Plagioclase ratios indicate that K, which is volatile, and Ga, which is volatile and siderophile, are depleted relative to chondrites. Ratios among nonvolatile elements confirm that the moon formed out of materials akin to chondritic meteorites.