Rare earth elements are commonly assumed to substitute only for Ca in clinopyroxene because of the similarity of ionic radii for REE 3+ and Ca 2+ in 8-fold coordination. The assumption is valid for Mg-rich clinopyroxenes for which observed mineral/melt partition coefficients are readily predicted by the lattice strain model for substitution onto a single site (e.g. Wood and Blundy, 1997). We show that natural Fe-rich pyroxenes in both silica-undersaturated and -oversaturated magmatic systems deviate from this behavior. Salites (Mg# 48 to 59) in phonolites from Tenerife, ferrohedenbergites (Mg# 14.2 to 16.2) from the rhyolitic Bandelier Tuff, and ferroaugites (Mg# 9.6 to 32) from the rhyolitic Rattlesnake Tuff have higher heavy REE contents than predicted by single-site substitution. The ionic radius of Fe 2+ in 6-fold coordination is substantially greater than that of Mg 2+ , hence we propose that, in Fe-rich clinopyroxenes, heavy REE are significantly partitioned between 8-fold Ca sites and 6-fold Fe-Mg sites such that Yb and Lu exist dominantly in 6-fold coordination. We also outline a REE-based method of identifying pyroxene-melt pairs in systems with multiple liquid and crystal populations, based upon the assumption that LREE and MREE reside exclusively in 8-fold coordination in pyroxene.Contrary to expectations, interpolation of mineral/melt partition coefficient data for heavy REE does not predict the behavior of Y. We speculate that mass fractionation effects play a role in mineral/melt lithophile trace element partitioning that is detectable among pairs of isovalent elements with near-identical radii, such as Y-Ho, Zr-Hf, and Nb-Ta.