Abstract: An improved genetic algorithm was applied and examined to optimize the weights of a neural network model for estimating root length density (RLD) distributions of winter wheat under salinity stress. Thereafter, soil water and solute transport with root-water-uptake in a soil-wheat system was simulated numerically, in which the estimated RLD distributions were incorporated. The results showed that the estimated RLD distributions of winter wheat using the neural network model combined with the improved genetic algorithm, as well as the simulated soil water content and salinity distributions, were comparably in agreement with the experimental data. The method can be used in modeling flow and transport under salinity or saline water irrigated areas.