Because of the difficulty and time involved in making exhaustive measurements of the geometric parameters of large tree crowns, simplifying hypotheses are often used in 3D virtual plant modelling, but the effects on the radiation balance of each approximation are rarely assessed. Three hybrid walnut trees aged 7-9 years were digitized to analyse the effect of the crown geometric variables on light capture. The six studied variables were: (1) leaf area, (2) number of leaves per annual shoot, (3) position of leaves, (4) orientation of leaves, (5) leaflet inclination, and (6) lamina shape. For each variable, a sensitivity analysis compared a reference, based on observed values, with scenarios consisting of simplifying hypotheses. The total incident light intercepted during a bright day and the distributions of leaf irradiance were calculated using the Archimed radiative transfer model. Since some of the crown parameters were generated stochastically, the radiation simulations were repeated until results stabilised. Simplified models can be used to calculate with satisfactory results individual leaf area and number of leaves per shoot. Conversely, differentiating statistical distributions of individual leaf area between short and long shoots is more difficult and may generate errors up to 30%. Leaf clumping is a determining factor and requires correct grouping of leaves around the annual shoots bearing them. The effect of position of leaves along the shoot is less than 2%. Simple statistical distributions are adequate for representing leaf angle. Finally, the effect of specific leaf geometry is very important, but it can be approached using a limited number of representative leaf shapes.