Abstarct: In this work, a quantitative structure-property relationship (QSPR) study was employed to predict the density of a diverse set of organic compounds including aliphatic and aromatic hydrocarbons, alcohols, amines, esters, ethers, carboxylic acids, as well as ketones over a wide range of temperatures and pressures. Two types of descxriptor selection method have been used in this study. In the first selection method, after analyzing the chemical structure of all studied compounds in this work, 24 functional groups were recognized baxsed on group contribution method (GCM). In the second method, the 3224 descxriptors were calculated using Dragon software for each of the studied organic compounds. Then, a subset of calculated descxriptors were selected from 22 classes of Dragon descxriptors with a genetic algorithm baxsed on partial least square (GA-PLS) method as a feature selection technique. Only 7 descxriptors were obtained by genetic algorithm (GA) as the most feasible descxriptors. The selected descxriptors with two feature selection methods and 2 experimental variables (temperature and pressure) as well as molecular weight of compounds were used as input nodes for constructing different artificial neural networks (ANNs). 4 feed-forward artificial neural network models were designed by combination of two training algorithms namely, Levenberg-Marquardt (LM) and Bayesian regularized (BR), with two transfer functions (logarithm-sigmoid and tangent-sigmoid). After training each of these models, the ANNs parameters such as, the number of the neurons in input and hidden laxyers, and the number of the iteration (epoch) have been optimized baxsed on the minimum value of the mean square error (MSE) for prediction of validation data set. Then, the performance of the selected optimized models was investigated by the internal and external test sets. The MSE and the determination coefficient (R2) for the internal test set using optimized GCM-ANN model were 18.15 and 0.9990, and those for optimized GA-ANN model were 22.92 and 0.9988, respectively. Similarly, the values of these parameters for external test set were obtained 96.74 and 0.9899 using GCM-ANN, and 1721.71 and 0.8622 using GA-ANN, respectively. The results showed that the optimized GCM-ANN model may be simulated the relationship between the structural descxriptors and density of organic compounds more accurate than the optimized GA-ANN model.