An architectural model is proposed to simulate 3D leaf area distribution for contrasting ( Vitis vinifera) vine training systems. Operating at the shoot level, the model describes stochastically prospected shoot volume (8 parameters) and leaf orientation (2 parameters) using distribution laws. Mean shoot leaf area is an input of the model. Parameterization was led for four genotype * training system (G * T) combinations. Validation was undertaken at two levels. At the canopy level, observed and simulated radiation interception efficiencies were compared. At the organ level, leaf irradiance distributions of simulated and 3D-described sample vines were evaluated. At both scales, observed and simulated values were closely correlated. The model tended to systematically overestimate self-shading into the canopies. However, its ability to discriminate and correctly rank the different G * T pairs was confirmed.