Serpentine banded veins are frequently observed in massive serpentinized peridotites. They form by extension or extensional shearing during hydrothermal alteration of peridotites. Serpentine minerals display different structural varieties, the occurrences of which are not well defined in terms of temperature, pressure, and chemistry, but may be controlled by departure from equilibrium and by the local water/rock ratio. Serpentines are therefore potential markers of environmental conditions during vein formation. However, they have never been used to assess the mechanism of banded vein formation. Using multi-scale microscopy techniques, and comparing detailed observations of natural samples from cm to nm scale with available experimental results, we attempt to deduce constraints on growth mechanisms of serpentines in banded veins. The banded internal structure and the filling along the vein-wall contact suggest a crack-seal mechanism of formation. Each crack is homogeneously filled with chrysotile and some rare polygonal serpentines (tubular serpentine varieties) and disseminated patches of gel-type protoserpentine. The tubes are not parallel to each other, but clearly show a preferred orientation perpendicular to the crack wall. Recent synthesis experiments describe a temporal succession of occurrence of these three serpentine microstructures. The observations suggest that such an evolution can occur in natural samples. The geometric peculiarities of macroscopic growth mechanisms in microscopic interstices may account for capillary effects. Based on this consideration, a simple qualitative model of serpentine banded vein formation is proposed. This model provides a possible origin for the enhancement and maintenance of a diffusional mass transfer from the matrix to the crack. This model also predicts the very good tracking of vein opening directions in such veins.