Fractional crystallization is a widely accepted mechanism to account for magmatic differentiation, but the location and geometry of the locations where this process takes place are still largely unknown. This is mainly due to the scarcity of occurrences where direct links between deep-seated cumulate rocks and their associated differentiation products are observed. The island of Fuerteventura in the Canary Archipelago offers such an opportunity. A dismembered Miocene volcano reveals a complex plutonic root zone, named PX1, consisting of a network of feeder conduits that can be related to remnants of contemporaneous lava flows. The lavas reveal a typical alkaline differentiation trend, whereas PX1 is composed of vertically layered cumulate lithologies ranging from olivine-rich wehrlites to clinopyroxenites and gabbros. Clinopyroxene phenocrysts in lavas (73·2–86·5 Mg#) exhibit the same zoning features and compositional ranges in major and trace elements as the plutonic clinopyroxenes, suggesting a direct genetic link between plutonic and effusive rocks. We interpret the PX1 rocks as resulting from the accumulation in volcanic conduits of phenocrysts segregated from ascending crystal-bearing magmas. Semi-quantitative modeling shows that fractional crystallization can indeed explain the evolution of the erupted lavas from basalt to basaltic trachyandesite. Progressive fractionation of the mineral assemblages olivine → olivine + minor clinopyroxene → clinopyroxene → clinopyroxene + plagioclase → plagioclase + kaersutite + apatite, with variable amounts of Fe–Ti oxides, can reproduce the major element compositional trend revealed by the Fuerteventura lavas. This calculated sequence coincides with the sequence of crystallization deduced from the textures of the PX1 rocks and could a priori generate the mineral assemblages that constitute the plutonic cumulates. However, complex core–mantle chemical zoning observed in PX1 clinopyroxenes suggests more complicated differentiation processes than simple crystal segregation and accumulation in vertical conduits. Primitive cores in PX1 clinopyroxenes (Mg# 80–88) display resorption features and sieve textures; they are interpreted as pre-existing crystals entrained from deeper levels or crystallized during the early stages of magma evolution. Clinopyroxene mantles display more evolved compositions (Mg# 74–80) with reverse zoning and external resorption features. Asymmetry in chemical zoning suggests crystallization in a confined environment such as a crystal mush. All these observations point to progressive mineral coarsening in the crystal cumulates, induced by the pervasive percolation of uprising melts on their way to the surface. The main factors controlling the variability of the cumulate lithologies are the crystal–melt segregation efficiency, the degree of differentiation and modal proportions of the magmas at their time of intrusion, and the extraction efficiency of the residual melts from the cumulates. The proposed model suggests that a differentiation process could take place in the feeder conduits and does not require storage of significant magma volumes (or magma chambers) in the crust to explain the observed evolution from mafic to intermediate magmas. Nevertheless, the existence of such large magma reservoirs is not excluded, in particular to explain the generation of significant volumes of differentiated lavas, as observed in other Canary Islands. Magma differentiation in vertical conduits opens up new perspectives in the understanding of fractional crystallization processes associated with the evolution of alkaline mafic magmas in ocean islands and continental intraplate settings.