Between 1927 and 2010, more than one hundred eruptions of Piton de la Fournaise produced ~1 km3 of lava, and the volcano’s summit collapsed twice (in 1931 and 2007). These lavas display, respectively, 20 and 65 % of the Sr–Nd and the Pb isotope ranges reported for La Réunion volcanoes over their known eruptive record (3.8 Ma). Variations in major and trace element concentrations and Sr–Pb isotopes do not define a temporal trend at the scale of the century, but display systematic short-term cyclic fluctuations. The positive correlation between 87Sr/86Sr and ratios of trace elements that are more versus less incompatible during partial melting of the mantle (e.g., Nd/Sm, La/Sm) probably results from the sampling of small-scale heterogeneities within the plume source. Changes in the degree of melting and/or crystallization are debated, but these appear ultimately linked to source properties. Lead isotopes do not co-vary with Sr isotopes, in part because of the partitioning of Pb into dense metallic phases that are preferentially sampled during high-flux eruptions. Taken together, Sr–Nd–Pb–Os–Th isotopes do not support contamination of magma with genetically unrelated components, such as the underlying Indian oceanic crust, mantle lithosphere, seawater, or seawater-altered lavas. Yet, in some rare cases (e.g. the 1998 Hudson eruption), the compositional patterns suggest that the parental magma assimilated older volcanic products within the edifice, such as crystal cumulates and/or interstitial differentiated melts. The geochemical fluctuations over the 1927–2010 time period constrain the residence time of magma in the shallow reservoir to 10–30 years and its size to 0.1–0.3 km3. The magma residence time during the course of the long-lived 1998 eruption is estimated to be an order of magnitude shorter, but the reservoir was probably of similar size. Instead, the shorter magma residence for the 1998 eruption was probably due to a higher magma flux.