Detailed study of a 65 cm harzburgite section perpendicular to an amphibole pyroxenite vein from the Lherz massif reveals a strong mineralogical and chemical zonation with distance from the vein-host boundary. At less than ∽ 20 cm, the host peridotite is modally metasomatized and displays patterns of increasing Fe, Ti, Mn, Al, Ca, Na, and HREE, and decreasing Mg and Ni toward the vein contact This zone is also relatively impoverished in Cr but is enriched in K and Sr. It is characterized by relatively unfractionated, mainly convex-upward, chondrite-normalized REE patterns. At a distance over ∽ 20 cm, the host peridotite displays the typical feature of cryptic metasomatism, i.e., selective LREE enrichment in otherwise anhydrous mineralogy. The chondrite-normalized REE patterns vary from U-shaped in the range 15–25 cm to strongly fractionated in the range 25–65 cm. These variations encompass the whole range reported from metasomatized peridotite nodules in alkali basalts. They may be accounted for by a single, silicate-melt, metasomatic event associated with infiltration of the Pyrenean alkali basalts into the most refractory peridotites, during their ascent through the subcontinental lithosphere, ∽ 100 Ma ago. The proposed model involves a chemical evolution of the infiltrated melt with increasing distance in the host. At < 15–20 cm, the melt composition would be strongly influenced by the proximity of the vein conduit, because of the existence of advective chemical fluxes through grain boundaries (short-range porous flow; distance of percolation < 1 m) and into small branching cracks, and a possibly dominant diffusive flux within the infiltrated melt. This may explain the reactivity of the melt towards the anhydrous peridotite mineralogy, the existence of chemical gradients for most elements, and the lack of REE chromato-graphic fractionation. At a distance of > 20–25 cm, chemical exchange with the conduit would be negligible and the melt composition would be mainly controlled by re-equilibration of the peridotite matrix during long-range (> 1 m) porous flow percolation. Thus the melt would be buffered by the amphibole peridotite mineralogy, except for LREE. This may explain the lack of mineralogical reaction and chemical enrichments (except for REE) in this zone, and the chromatographic fractionation of REE. We propose a quantitative model of diffusion and percolation-controlled metasomatism associated with infiltration of alkali basalts into peridotites hosting vein-conduits. We also suggest that silicate-melt percolation may explain mineral disequilibrium features observed in mantle xenoliths.