New major-element, trace-element, and isotopic (Nd, Sr) analyses of undersaturated alkaline lavas from the Kilimanjaro volcano (north Tanzania) are presented. These data concern 54 samples, ranging from basanites to phonolites, collected during a 1 mo field trip in March 2005. The three main cones of Kilimanjaro were sampled, Shira, Mawenzi, and Kibo, together with numerous parasitic cones located on a SE lineament on the main edifice. On the basis of both spatial distribution and major- and trace-element characteristics of analyzed samples, the previous classification of Kilimanjaro lavas is simplified into five groups: Shira, Mawenzi, Kibo 1, Kibo 2, and parasitic activity, each of which has distinct petrological and geochemical features. The major- and trace-element characteristics of the rare primitive lavas erupted on the volcano yield the ubiquitous signature of amphibole within the magma source. We propose that Kilimanjaro melts originated from the partial melting of lithospheric mantle. Combined modeling of trace-element behavior during partial melting + fractional crystallization and isotopic constraints allow us to propose a schematic model of melt genesis under the Kilimanjaro area. Thermal heating of the ancient continental lithosphere by an upwelling plume triggered partial melting in parts of the lithosphere where amphibole was present and led to the Shira volcanic episode. Then, during a time span of ~1 m.y., the depleted lithosphere was progressively infiltrated by plume melts that resulted in crystallization of a new generation of metasomatic amphibole. Finally, this rejuvenated lithosphere underwent partial melting, leading to the magmas that formed the main edifice (Mawenzi and Kibo), and leading to a progressive depletion of the source with time. An active contribution of true asthenospheric melts during the last magmatic events of the volcano cannot be excluded, but further detailed isotopic investigations are needed to test the model.