Mathematical models that incorporate biotic and abiotic attributes are important tools for improving fertilizer use efficiency and reducing production costs for soybean [Glycine max (L.) Merrill] crop. In this study, artificial neural networks (ANNs) were used to estimate soybean grain yield (GY) under subtropical conditions in Brazil from plant morphological and nutritional data collected from 16 cultivars in two growing seasons. The ANNs were adequately trained, with a mean squared error of approximately 10−5 between the outputs obtained (via ANN) and desired (via experimental field), equivalent to a mean percentage error of 70.1 kg ha−1 (1.6%), confirming their efficacy as a tool to estimate GY. Smaller plant height, higher foliar calcium, magnesium and chlorophyll concentrations, and greater numbers of grains per pod and branches per plant were associated with higher GY, whereas oil content, crude protein content, and foliar manganese and potassium concentrations had no predicted effects on GY.