Leaf Area Index (LAI) is a key input for many crop models. The LAI patterns measured in situ are time consuming and labor intensive and could be substituted by intelligent techniques of approximation as artificial neural networks (ANNs). The objective of this study was to evaluate the possibility to estimate the evolution of LAI and height of maize canopies in southeastern of Buenos Aires Province, Argentine using neural network models. A field experiment under non limiting condition was carried out to generate a range of environmental conditions (four planting dates and three hybrids with contrasting maturity). Periodical measurements of LAI on tagged plants were used to develop, evaluate and test the neural networks to approximate variation of leaf area index and height at plot scale. Data from canopy structure properties as leaf area, height and leaf area density profile were obtained by non-destructive methods. Planting date (PD), relative maturity of the hybrid (MR) and thermal time from emergence (TTE) were the inputs to the ANNs. A decomposition method based on Garson’s algorithm was applied to quantify the relative importance for each input variable. The method provides a better description of the knowledge learned by the networks during the training process. Sensitivity analysis was performed to identify relevant variables and quantify the risk of a given combination of variables. The RM showed a major contribution in ANNs to estimate LAI and HLL. Both trained ANNs were most sensitive to TTE than the remaining inputs.Sociedad Argentina de Informática e Investigación Operativa (SADIO)