Rosetta program follows a hierarchical approach to estimate the hydraulic parameters of soils using five levels of input data. The first level consists of a look up table containing the average hydraulic parameters for each soil textural class, but this level is avoided because its low accuracy. The other four levels are based on neural network analysis. These levels were used to predict soil hydraulic parameters and water retention of different soil samples. Generally, the sensitivity analysis (MSE) showed that soil particle size distribution had a major influence on the shape of water retention curve, while bulk density, soil water content at both 33kPa and 1500 kPa had increased the accuracy of the program. These increments of accuracy were differ from soil sample to another. In some cases, particle size distribution was enough to significantly predict soil water retention and more precise than in the case of adding the other parameters, such as soil bulk density or soil water contents  at both  33 kPa and 1500 kPa pressure head to the used  program as input parameters.