We describe the field relations, petrographic, mineralogical and geochemical characteristics of an exceptional “golden” pumice belonging to a tephra layer exposed on the summit area of Stromboli volcano, Italy. Pumice sample PST-9 comes from a fallout deposit older than a spatter agglutinate sequence emplaced during the twentieth century. The eruption that produced it had a size exceeding that of intermediate paroxysms but was smaller than large-scale, spatter-forming, paroxysms from the sixteenth century and 1930 A.D. Lapilli are strongly vesicular and crystal-poor, similar to other “golden” pumices. Modal proportions include 89 vol% glass, 8 vol% clinopyroxene, 1–2 vol% olivine and 1–2 vol% plagioclase. Plagioclase is represented by reacted crystals coming from the shallow resident magma and incorporated in the pumice during eruption. A total of 74 and 44 crystals of olivine and clinopyroxene, respectively, were examined and 187 and 99 electron microprobe analyses obtained. Fo in olivine ranges between 70 and 92 mol% and Fs in clinopyroxene between 3 and 13 mol%. PST-9 hosts a higher proportion of Fo-rich olivine and Fs-poor clinopyroxene than the other “golden” pumices. Groundmass glasses are basaltic (Mg# = 66–69), as are most rim glasses around olivine and clinopyroxene, and glass inclusions in clinopyroxene. They are more primitive than in the other “golden” pumices. A few rim glasses and glass inclusions are shoshonitic (Mg# = 45–50). Most glass inclusions in olivine have CaO/Al2O3 higher than the other glasses and the whole-rock. PST-9 has the highest bulk MgO, CaO, Mg# and CaO/Al2O3 and the lowest FeOt of all “golden” pumices analysed to date. Analysis of Fe-Mg partitioning between olivine, clinopyroxene and melt allows three crystallization stages to be recognized. The first involves primitive mantle-derived melts (Mg# = 74–80), the second basaltic melts represented by groundmass glasses and the third is associated with more evolved melts represented by the shoshonitic glasses. The population of crystals in “golden” pumices is heterogeneous not only because of crystal incorporation from the shallow resident magma, but also because of pre-eruptive recharge of the deep reservoir with primitive melts. Differences between PST-9 and the other “golden” pumices in terms of groundmass glass composition and distribution of olivine and clinopyroxene compositions reflect contrasted replenishment rates of the deep reservoir with primitive liquids. Gabbroic inclusions in a clinopyroxene crystal provide a direct illustration of melt wall-rock interaction and stress the variability of the deep reservoir in terms of temperature, crystallinity and phase assemblages. Deep crystallization of plagioclase should be considered as a possibility at Stromboli. PST-9 is exceptionally well representative of the early magmatic evolution of “golden” pumices.