Modelling and predicting the hydrological processes associated with discontinuous, perennial crop growing areas, like the French vine-cultivated Mediterranean region, requires detailed quantitative information on spatial crop structure. A method was developed to provide such information by spatial frequency analysis on very high spatial resolution data. A simple crop geometry model, based on general knowledge and field observations was applied to the Fourier power spectrum of aerial colour imagery obtained over the La Peyne valley (Herault, France). This method, applied on a per-field basis, using a digital land register map, allowed the identification of vineyards and characterization of their crop spacing, orientation and training mode. Results showed good performances of this procedure under all conditions encountered, i.e. very variable soil surface optical properties and spatial structure. The main vineyard training modes, goblet and wire-trained, as well as orchards and continuous crop/fallow fields were well classified. Only non-standard, intermediate vine training modes were not always correctly recognized. Additionally, the frequency analysis provided, may be useful to environmental studies. The geometrical accuracy of image warping and minimal field size and width required by the method are discussed.