We report here the most complete dataset for major and trace elements, as well as Sr isotopic compositions, of magmas erupted by Stromboli since the onset of present-day activity 1,800 years ago. Our data relate to both porphyritic scoria and lava originating in the uppermost parts of the feeding system, plus crystal-poor pumice produced by paroxysmal explosive eruption of deep-seated, fast ascending, magma. The geochemical variations recorded by Stromboli's products allow us to identify changes in magma dynamics affecting the entire plumbing system. Deep-seated magmas vary in composition between two end-members having different key ratios in strongly incompatible trace elements and Sr isotopes. These features may be ascribed to mantle source processes (fluid/melt enrichment, variable degrees of melting) and occasional contamination by deep, mafic, cumulates. Temporal trends reveal three phases during which magmas with distinct geochemical signatures were erupted. The first phase occurred between the third and fourteenth centuries AD and was characterised by the eruption of evolved magmas sharing geochemical and Sr isotopic compositions similar to those of earlier periods of activity (<12 ka-Neostromboli and San Bartolo). The second phase, which began in the sixteenth century and lasted until the first half of the twentieth century, produced more primitive, less radiogenic, magmas with the lowest Ba/La and Rb/ Th ratios of our dataset. The last phase is ongoing and is marked by a magma having the lowest Sr isotopic composition and highest Rb/Th ratio of the dataset. While this new magma can be clearly identified in the pumice erupted during the last two paroxysmal eruptions of 2003 and 2007, shallow degassed magma extruded during this time span records significant geochemical and isotopic heterogeneities. We thus suggest that the shallow reservoir has been only partially homogenised by this new magma influx. We conclude that compositional variations within the shallow magma system of a persistently active volcano provide only a biassed signal of ongoing geochemical changes induced by deep magma refilling. We argue that source changes can only be identified by interpreting the geochemistry of pumice, because it reliably represents magma transferred directly from deep portions of the plumbing system to the surface.