We have analyzed the major- and trace-element contents of minerals and whole rocks for two peridotite xenoliths from SE Australia (one fertile, the other infertile) using both solution chemistry (ICP–MS) and microbeam techniques (EMP, LA–ICP–MS). The incompatible lithophile trace elements are contained mainly in solid mineral phases at upper-mantle conditions, with no significant concentrations occurring on grain boundaries or in fluid inclusions. No exotic accessory minerals are required hosts for elements such as Nb and Ta. Mass balance is achieved for the highly incompatible trace elements only with inclusion of data for precursor amphibole, now present as pockets of alkali-, alumina- and silica-rich glass containing euhedral olivine, clinopyroxene, and spinel daughter crystals. The glass in these pockets is enriched in incompatible trace elements, particularly Ba, Nb and Ta, reflecting the preferential partitioning of these elements into amphibole at upper-mantle conditions. Orthopyroxene–clinopyroxene partition coefficients are similar for most incompatible elements (except Ti) between the two peridotites, suggesting that compositional differences do not significantly affect subsolidus partitioning for these elements. Comparison of our data with those from the literature reveals large variations in Dopx/cpx for all incompatible trace elements. For the REE, Zr (Hf) and Sr, these variations correlate with reciprocal equilibration temperature, reflecting the substitution of these elements for Ca in the M2 site. In contrast, poor correlation exists for V and no correlations exist for Ti and Nb between Dopx/cpx and temperature, which may reflect their substitution into multiple sites in the pyroxene structure and the influence of mineral composition on their partitioning. These features highlight the need for caution in inferring crystal–liquid D-values from subsolidus partitioning relations.